Unreal Speech

API Docs Pricing Studio Blog High Volume Inquiry Sign In Fast & Affordable Text-to-Speech API 11x cheaper than Eleven Labs Stream audio in 300ms Request up to 10-hour audio Includes per-word timestamps Get a Free API Key → Live Demo Non-FictionFictionNewsBlogConversation Amid the intricate labyrinth of human neurons lies a molecule that has confounded and fascinated scientists for ages: the neurotransmitter known as dopamine. Often heralded as the pleasure molecule, dopamine's role is far more nuanced than just mediating euphoria. 0/250 Voice Sierra AutumnMelodyHannahEmilyIvyKaitlynLunaWillowLaurenSierraNoahJasperCalebRonanEthanDanielZaneRowan Language 🇺🇸 US English 🇺🇸 US English🇬🇧 UK English🇨🇳 Mandarin Chinese🇮🇳 Hindi🇪🇸 Spanish🇧🇷 Portuguese🇯🇵 Japanese🇫🇷 French🇮🇹 Italian Format Standard StandardMP3Phone CallPCM µ-law Normal Speed 1.0 Pitch Bitrate 192k 192k128k96k64k48k32k16k Synthesize Speed 0 s Filesize 0 kb Per-word Timestamps Highlight words in sync with the speech. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. How To Get Timestamps NEW: Use a websocket connection to stream both audio and timestamps using /streamWithTimestamps Use /speech or /synthesisTasks with TimestampType set to word or sentence (see docs). The response will have a TimestampsUri containing JSON like this: [ { "word": "Our", "start": 0.5, "end": 0.6666666666666667, "text_offset": 0 }, { "word": "model", "start": 0.6666666666666667, "end": 1.075, "text_offset": 4 }, { "word": "achieves", "start": 1.075, "end": 1.5166666666666666, "text_offset": 10 }, ... ] API Docs → Demo Code → The more you use it, the cheaper it gets Start for free. Stay for discounts. Number of Characters Per month 625M Audio Duration Estimated ~14K hours Enterprise Plan Includes 625M characters $4999 a month High Volume Inquiry Additional Usage $8 per 1M characters — Comparison vs Amazon AmazonMicrosoftGoogleElevenLabsPlay.ht $4999 a month $10.0K a month Scarlett ScarlettDan Score Fiction 4.72 Non-Fiction 4.37 Conversation 3.91 Neural NeuralStandard Score Fiction 3.00 Non-Fiction 2.51 Conversation 2.63 Price Comparison Unreal ElevenLabs Play.ht Amazon Microsoft Google Calculated using public prices. Custom plans may be cheaper. * 1 minute of audio = roughly 750 characters (~150 WPM) Get Started for Free → Need a custom solution? Contact us "Unreal Speech saved us 75% on our text-to-speech cost. It sounds better than Amazon Polly, and is much cheaper. We switched over at high volumes, and often processing 10,000+ pages per hour. Unreal was able to handle the volume, while delivering a high quality listening experience." Derek Pankaew — CEO, Listening.com 7B Characters per month 0.3s Latency 99.9% Uptime Code Samples Get started quickly with our simple text-to-speech API. /stream /speech /synthesisTasks /streamWithTimestamps SDK Python PythonNode.jsReact NativeBash Copy # Endpoint: /stream # - Convert up to 1,000 characters ASAP # - Synchronous, instant response (0.3s) # - Streams back raw audio data (no timestamps) import requests response = requests.post( 'https://api.v8.unrealspeech.com/stream', headers = { 'Authorization' : 'Bearer YOUR_API_KEY' }, json = { 'Text': '''''', # Up to 1,000 characters 'VoiceId': '', # af, af_bella, af_sarah, am_adam, am_michael, bf_emma, bf_isabella, bm_george, bm_lewis, af_nicole, af_sky 'Bitrate': '192k', # 320k, 256k, 192k, ... 'Speed': '0', # -1.0 to 1.0 'Pitch': '1', # 0.5 to 1.5 'Codec': 'libmp3lame', # libmp3lame or pcm_mulaw } ) with open('audio.mp3', 'wb') as f: f.write(response.content) # Endpoint: /speech # - Up to 3,000 characters # - Synchronous, takes ~1s per 700 chars # - Returns MP3 and JSON timestamp URLs import requests response = requests.post( 'https://api.v8.unrealspeech.com/speech', headers = { 'Authorization' : 'Bearer YOUR_API_KEY' }, json = { 'Text': '''''', # Up to 3,000 characters 'VoiceId': '', # Scarlett, Dan, Liv, Will, Amy 'Bitrate': '192k', # 320k, 256k, 192k, ... 'Speed': '0', # -1.0 to 1.0 'Pitch': '1', # 0.5 to 1.5 'TimestampType': 'sentence' # word or sentence } ) print(response.json()) # Endpoint: /synthesisTasks # - Up to 500,000 characters # - Asynchronous, takes ~1s per 800 chars # - Returns a TaskId (use it to check status) import requests response = requests.post( 'https://api.v8.unrealspeech.com/synthesisTasks', headers = { 'Authorization' : 'Bearer YOUR_API_KEY' }, json = { 'Text': '''''', # Up to 500,000 characters 'VoiceId': '', # Scarlett, Dan, Liv, Will, Amy 'Bitrate': '192k', # 320k, 256k, 192k, ... 'Speed': '0', # -1.0 to 1.0 'Pitch': '1', # 0.5 to 1.5 'TimestampType': 'sentence', # word or sentence #'CallbackUrl': '', # pinged when ready } ) print(response.json()) # Endpoint: /streamWithTimestamps # - Convert text with real-time word-level timestamps # - Streams audio with precise word timing information # - Perfect for word-by-word highlighting import websocket import json audio_chunks = [] timestamps = [] def on_message(ws, message): try: data = json.loads(message) if data.get("type") == "progress" and "message" in data: msg_val = data["message"] if isinstance(msg_val, list): timestamps.extend(msg_val) elif data.get("type") == "complete": ws.close() except Exception: # Assume binary audio data audio_chunks.append(message) def on_open(ws): config = { "Text": "", "VoiceId": "", "Model": "kokoro", "Codec": "libmp3lame", "SampleRate": 24000, "Speed": 0, "Bitrate": "192k", "Pitch": 1.0, "TimestampType": "word" } ws.send(json.dumps(config)) ws = websocket.WebSocketApp( "wss://api.v8.unrealspeech.com/streamWithTimestamps", on_open=on_open, on_message=on_message ) ws.run_forever() // Endpoint: streamWithTimestamps (WebSocket) # - Convert text with real-time word-level timestamps # - Streams audio with precise word timing information # - Perfect for word-by-word highlighting import websocket import json import threading import time # Files to store output audio_file = 'output.mp3' timestamps_file = 'timestamps.json' audio_chunks = [] timestamps = [] def on_message(ws, message): if isinstance(message, str): # Handle JSON messages (timestamps or status) data = json.loads(message) if data.get('type') == 'progress' and 'message' in data: # Add timestamp information msg_val = data['message'] if isinstance(msg_val, list): timestamps.extend(msg_val) elif data.get('type') == 'complete': # Save the audio and timestamps when complete with open(audio_file, 'wb') as f: import io buffer = io.BytesIO() for chunk in audio_chunks: buffer.write(chunk) f.write(buffer.getvalue()) # Save timestamps (filter out empty ones) filtered = [ts for ts in timestamps if ts and 'word' in ts] with open(timestamps_file, 'w') as f: json.dump(filtered, f, indent=2) print("Audio saved to:", audio_file) print("Timestamps saved to:", timestamps_file) ws.close() else: # Handle binary audio data audio_chunks.append(message) def on_open(ws): # Send configuration when connection opens config = { 'Text': 'Welcome to Unreal Speech.', 'VoiceId': '', 'Model': 'kokoro', 'Codec': 'libmp3lame', 'SampleRate': 24000, 'Bitrate': '192k', 'Pitch': 1.0, 'TimestampType': 'word' } ws.send(json.dumps(config)) # Connect to WebSocket endpoint websocket.enableTrace(False) ws = websocket.WebSocketApp( "wss://api.v8.unrealspeech.com/streamWithTimestamps", header={'Authorization': 'Bearer YOUR_API_KEY'}, on_open=on_open, on_message=on_message ) # Run WebSocket connection ws.run_forever() // Short endpoint: /stream // - Up to 1,000 characters // - Synchronous, instant response (0.3s+) // - Streams back raw audio data const axios = require('axios'); const fs = require('fs'); const headers = { 'Authorization': 'Bearer YOUR_API_KEY', }; const data = { 'Text': '', // Up to 1,000 characters 'VoiceId': '', // Scarlett, Dan, Liv, Will, Amy 'Bitrate': '192k', // 320k, 256k, 192k, ... 'Speed': '0', // -1.0 to 1.0 'Pitch': '1', // 0.5 to 1.5 'Codec': 'libmp3lame', // libmp3lame or pcm_mulaw }; axios({ method: 'post', url: 'https://api.v8.unrealspeech.com/stream', headers: headers, data: data, responseType: 'stream' }).then(function (response) { response.data.pipe(fs.createWriteStream('audio.mp3')) }); // Medium endpoint: /speech // - Up to 3,000 characters // - Synchronous, takes ~1s per 700 chars // - Returns MP3 and JSON timestamp URLs const axios = require('axios'); const headers = { 'Authorization': 'Bearer YOUR_API_KEY', }; const data = { 'Text': '', // Up to 3,000 characters 'VoiceId': '', // Scarlett, Dan, Liv, Will, Amy 'Bitrate': '192k', // 320k, 256k, 192k, ... 'Speed': '0', // -1.0 to 1.0 'Pitch': '1', // 0.5 to 1.5 'TimestampType': 'sentence', // word or sentence }; axios({ method: 'post', url: 'https://api.v8.unrealspeech.com/speech', headers: headers, data: data, }).then(function (response) { console.log(JSON.stringify(response.data)); }); // Long endpoint: /synthesisTasks // - Up to 500,000 characters // - Asynchronous, takes ~1s per 800 chars // - Returns a TaskId (use to check status) const axios = require('axios'); const headers = { 'Authorization': 'Bearer YOUR_API_KEY', }; const data = { 'Text': '', // Up to 500,000 characters 'VoiceId': '', // Scarlett, Dan, Liv, Will, Amy 'Bitrate': '192k', // 320k, 256k, 192k, ... 'Speed': '0', // -1.0 to 1.0 'Pitch': '1', // 0.5 to 1.5 'TimestampType': 'sentence', // word or sentence //'CallbackUrl': '', // pinged when ready }; axios({ method: 'post', url: 'https://api.v8.unrealspeech.com/synthesisTasks', headers: headers, data: data, }).then(function (response) { console.log(JSON.stringify(response.data)); }); Sample Response { 'SynthesisTask': { 'CreationTime': '2023-09-01T15:05:22.15Z', 'OutputUri': 'https://unreal-tts-live-demo.s3-us-west-1.amazonaws.com/d8ef514d.mp3', 'RequestCharacters': 14, 'TaskId': 'd8ef514d', 'TaskStatus': 'scheduled', 'TimestampsUri': 'https://unreal-tts-live-demo.s3-us-west-1.amazonaws.com/d8ef514d.json', 'VoiceId': 'Scarlett' } } Check Task Status Make a GET request with TaskId in the URL to check the status. const axios = require('axios'); const headers = { 'Authorization': 'Bearer YOUR_API_KEY', }; axios({ method: 'get', url: 'https://api.v8.unrealspeech.com/synthesisTasks/d8ef514d', headers: headers, }).then(function (response) { console.log(JSON.stringify(response.data)); }); // Short endpoint: /stream // - Up to 1,000 characters // - Synchronous, instant response (0.3s+) // - Streams back raw audio data import React from "react"; import { View, Button, ActivityIndicator } from "react-native"; import { useUnrealSpeech, blobToDataURI } from "react-native-unrealspeech"; import { Audio } from "expo-av"; export default function App() { const { stream, requestState } = useUnrealSpeech("YOUR_API_KEY"); const play = async (response: any) => { try { if (!response.ok) { throw new Error("Network response was not ok"); } const blobData = await response.blob(); const blob = new Blob([blobData], { type: "audio/mp3" }); const uri = await blobToDataURI(blob); const { sound } = await Audio.Sound.createAsync({ uri }, {}); console.log(uri); await sound.playAsync(); } catch (error) { console.error("Error occurred while playing sound:", error); } }; const handleSpeak = async () => { try { const response = await stream( "", // Up to 1,000 characters "", // Scarlett, Dan, Liv, Will, Amy "192k", // 320k, 256k, 192k, ... 0, // Speed: -1.0 to 1.0 1, // Pitch: 0.5 to 1.5 "libmp3lame" // Codec: libmp3lame or pcm_mulaw ); play(response); } catch (error) { console.error("Error occurred while handling speak:", error); } }; return ( {requestState === "loading" ? ( ) : ( )} ); } // Medium endpoint: /speech // - Up to 3,000 characters // - Synchronous, takes ~1s per 700 chars // - Returns MP3 and JSON timestamp URLs import React from "react"; import { StatusBar } from "expo-status-bar"; import { StyleSheet, Button, View, ActivityIndicator } from "react-native"; import { useUnrealSpeech } from "react-native-unrealspeech"; export default function App() { const { speech, requestState } = useUnrealSpeech("YOUR_API_KEY"); const handleSpeak = async () => { const mySpeech = await speech( "", // Up to 3,000 characters "", // Scarlett, Dan, Liv, Will, Amy "192k", // 320k, 256k, 192k, ... "sentence" // word or sentence ); console.log(mySpeech); }; return ( {requestState === "loading" ? ( ) : ( )} ); } const styles = StyleSheet.create( { container: { flex: 1, backgroundColor: "#fff", alignItems: "center", justifyContent: "center" } } ); // Long endpoint: /synthesisTasks // - Up to 500,000 characters // - Asynchronous, takes ~1s per 800 chars // - Returns a TaskId (use to check status) import React from "react"; import { StatusBar } from "expo-status-bar"; import { StyleSheet, Button, View, ActivityIndicator } from "react-native"; import { useUnrealSpeech } from "react-native-unrealspeech"; export default function App() { const { createSynthesisTask, getSynthesisTaskStatus, speech, stream, status, requestState } = useUnrealSpeech("YOUR_API_KEY"); const handleSpeak = async () => { const task = await createSynthesisTask( "", // Up to 500,000 characters "", // Scarlett, Dan, Liv, Will, Amy "192k", // 320k, 256k, 192k, ... "sentence" // word or sentence ); if (status) { const response = await getSynthesisTaskStatus(status); console.log(response); } }; return ( {requestState === "loading" ? ( ) : ( )} ); } const styles = StyleSheet.create( { container: { flex: 1, backgroundColor: "#fff", alignItems: "center", justifyContent: "center" } } ); # Short endpoint: /stream # - Up to 1,000 characters # - Synchronous, instant response (0.3s+) # - Streams back raw audio data curl -X POST "https://api.v8.unrealspeech.com/stream" -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_API_KEY" --data '{"Text": "", "VoiceId": "", "Bitrate": "128k", "Speed": "0", "Pitch": "1", "Codec": "libmp3lame"}' --output audio.mp3 # Medium endpoint: /speech # - Up to 3,000 characters # - Synchronous, takes ~1s per 700 chars # - Returns MP3 and JSON timestamp URLs curl -X POST "https://api.v8.unrealspeech.com/speech" -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_API_KEY" --data '{"Text": "", "VoiceId": "", "Bitrate": "192k", "Speed": "0", "Pitch": "1", "TimestampType": "sentence"}' # Long endpoint: /synthesisTasks # - Up to 500,000 characters # - Asynchronous, takes ~1s per 800 chars # - Returns a TaskId (use to check status) curl -X POST "https://api.v8.unrealspeech.com/synthesisTasks" -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_API_KEY" --data '{"Text": "", "VoiceId": "", "Bitrate": "128k", "Speed": "0", "Pitch": "1", "TimestampType": "sentence"}' Sample Response { 'SynthesisTask': { 'CreationTime': '2023-09-01T15:05:22.15Z', 'OutputUri': 'https://unreal-tts-live-demo.s3-us-west-1.amazonaws.com/d8ef514d.mp3', 'RequestCharacters': 14, 'TaskId': 'd8ef514d', 'TaskStatus': 'scheduled', 'TimestampsUri': 'https://unreal-tts-live-demo.s3-us-west-1.amazonaws.com/d8ef514d.json', 'VoiceId': 'Scarlett' } } Check Task Status Make a GET request with TaskId in the URL to check the status. curl -X GET "https://api.v8.unrealspeech.com/synthesisTasks/d8ef514d" -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_API_KEY" API Docs → Ready to dive in? Get a Free API Key → Super Fast Text-to-Speech API Build real-time apps. Generate long-form audio fast. /stream Play audio in 0.3s /synthesisTasks Create 10-hour audio in 15 min Get Started for Free → FAQ Do you offer voices in other languages? Yes, we provide 48 voices across 8 different languages, including US English, UK English, Mandarin Chinese, Hindi, Spanish, Portuguese, Japanese, French and Italian. Can I create custom voices (voice cloning)? Not right now, but we're working on it! What happens if I use all of my monthly characters? Additional usage over the monthly allowance will be charged daily at the rate of your current plan: Basic – $16 per 1M characters Plus – $12 per 1M characters Pro – $10 per 1M characters Enterprise – $8 per 1M characters What happens to unused characters at the end of the month? Free plan – Characters are reset on the 1st of every month. Paid plan – Unused characters roll over to the next billing cycle. Can I use generated audio commercially? Yes, audio generated with Unreal Speech can be used commercially. The following terms apply, based on your subscription plan: Free plan – You must attribute Unreal Speech when publishing audio by including a link to "unrealspeech.com" in the description. Paid plan – You do not need to include any attribution. How do I update my payment method? Go to your Dashboard and choose "Manage Subscription". How do I cancel my subscription? You can cancel your subscription at any time. Go to your Dashboard and choose "Manage Subscription". Do you have an affiliate program? Yes! You can earn 15% recurring on all paid referrals. Click here to sign up. Made with ❤️ in SF Powered by Kokoro TTS API Docs Pricing Studio Contact Blog Compare Browse AI Apps Affiliate Recent Posts Integrating FastSpeech 2 for Text-to-Speech Synthesis with Fairseq and Hugging Face Exploring the Potential of GPT-SoVITS-Fork for Text-to-Speech Applications Exploring the GPT-SoVITS Kancolle Zuikaku TTS Model: A Comprehensive Guide Exploring Voice Synthesis with ESPnet: A Deep Dive into the kan-bayashi_csmsc_fastspeech Model Introducing OpenVoice: Revolutionizing Text-to-Speech with Instant Voice Cloning and Multilingual Capabilities How to Leverage Twelve Labs API for Effortless YouTube Video Summaries, Chapters, and Highlights API Docs Pricing Studio Blog Sign In Sign In Sign in with GoogleGoogleSign in with emailEmailBy continuing, you are indicating that you accept our Terms of Service and Privacy Policy. Copied to clipboard! --- API Docs Pricing Studio Blog Sign In Pricing that Scales with You Start for free. Stay for volume discounts. Free $0 250K characters 6 hours of audio Select Basic $49 $4.99/mo Discount for first 6 months 3M characters 67 hours of audio Select Plus $499/mo 42M characters 933 hours of audio Select Pro $1499/mo 150M characters 3K hours of audio Select Enterprise $4999/mo 625M characters 14K hours of audio Select Custom High Volume Inquiry 1B+ characters Volume discounts High Volume Inquiry * Audio duration is approximate * 1 minute of audio is roughly 750 characters Made with ❤️ in SF Powered by Kokoro TTS API Docs Pricing Studio Contact Blog Compare Browse AI Apps Affiliate Recent Posts Integrating FastSpeech 2 for Text-to-Speech Synthesis with Fairseq and Hugging Face Exploring the Potential of GPT-SoVITS-Fork for Text-to-Speech Applications Exploring the GPT-SoVITS Kancolle Zuikaku TTS Model: A Comprehensive Guide Exploring Voice Synthesis with ESPnet: A Deep Dive into the kan-bayashi_csmsc_fastspeech Model Introducing OpenVoice: Revolutionizing Text-to-Speech with Instant Voice Cloning and Multilingual Capabilities How to Leverage Twelve Labs API for Effortless YouTube Video Summaries, Chapters, and Highlights API Docs Pricing Studio Blog Sign In Sign In Sign in with GoogleGoogleSign in with emailEmailBy continuing, you are indicating that you accept our Terms of Service and Privacy Policy. Copied to clipboard! --- Unreal Speech StudioStudio Dark Mode 🇺🇸 US English Language 🇺🇸 US English🇬🇧 UK English🇨🇳 Mandarin🇮🇳 Hindi🇪🇸 Spanish🇧🇷 Portuguese🇯🇵 Japanese🇫🇷 French🇮🇹 Italian Chloe Female Welcome to Kokoro TTS Studio! This free, efficient text-to-speech application uses advanced AI to transform text into natural-sounding audio. You can type, paste, or edit text here. It offers 48 natural-sounding AI voices across 8 languages: American, British, Chinese, Hindi, Spanish, Portuguese, Japanese and French. For commercial use cases, please check out our API documentation. Clear 388 / 500 characters Generate Get API Key Your Voice-overs Your recordings will appear here. Kokoro TTS Studio: Free Online Text-to-Speech Demo Welcome to Kokoro TTS Studio powered by Unreal Speech - the ultimate playground for the revolutionary 82M parameter open-source text-to-speech engine! Simply type your text, choose from our extensive library of 48 natural-sounding voices across 8 languages, and instantly generate high-quality speech that rivals premium commercial services. Convert text to speech in your browser, right now and download your audio files for free. What is Kokoro TTS: The Free Open-Source Speech Generator Kokoro TTS is a groundbreaking open-source text-to-speech model that's revolutionizing the AI voice landscape. With a remarkably tiny footprint of just 82 million parameters (a fraction of what other models use), Kokoro delivers astonishingly natural speech synthesis that outperforms models 5-15× its size in both quality and speed. Why Kokoro TTS Is Making Waves in the AI Community Incredibly Compact Yet Powerful: At just 82M parameters, Kokoro is dramatically smaller than competing models like XTTS v2 (467M) and MetaVoice (1.2B), yet produces equal or better voice quality Lightning-Fast Performance: Generates speech up to 210× real-time on GPU and 3-11× real-time on CPU—making it perfect for real-time applications and batch processing Resource-Efficient Design: Runs smoothly on consumer hardware without requiring expensive cloud infrastructure or specialized equipment Truly Open Source & Free: Licensed under Apache 2.0, allowing both commercial and non-commercial use with no restrictions Award-Winning Quality: Achieved 1st place in the HuggingFace TTS Spaces Arena for single-speaker speech quality, beating models many times its size Multilingual Support: Speaks multiple languages fluently, including English (US/UK), French, Hindi, Spanish, Japanese, Chinese, Italian, and Portuguese Try the live demo above to experience the remarkable quality yourself! Simply type your text above, select a voice, and click "Generate" to hear Kokoro TTS in action. Explore 48 Voices Across 8 Languages Kokoro TTS Studio supports English (US/UK), French, Hindi, Spanish, Japanese, Chinese, Italian, and Portuguese. Browse our extensive library of realistic voices spanning multiple languages and accents. Each voice has been carefully trained to deliver natural intonation and clarity: Display Name Voice ID Language Gender Description Hannah af_bella English (US) Female American female with reserved personality Kaitlyn af_nicole English (US) Female American female with whisper-like voice, looks casual Lauren af_sarah English (US) Female American female, probably an educator; confident Sierra af_sky English (US) Female American female with high level of composure Noah am_adam English (US) Male American male with confident personality Daniel am_michael English (US) Male American male with confident personality Chloe bf_emma English (UK) Female British female Amelia bf_isabella English (UK) Female British female with calm personality Edward bm_george English (UK) Male British male, mature voice Oliver bm_lewis English (UK) Male British male with confident personality Élodie ff_siwis French Female Young French female voice Ananya hf_alpha Hindi Female Young Hindi female voice Priya hf_beta Hindi Female Young Hindi female voice Arjun hm_omega Hindi Male Young Hindi male voice Sakura jf_alpha Japanese Female Young Japanese female voice Hana jf_gongitsune Japanese Female Young Japanese female voice Haruto jm_kumo Japanese Male Young Japanese male voice Lucía ef_dora Spanish Female Young Spanish female voice Mateo em_alex Spanish Male Young Spanish male voice Giulia if_sara Italian Female Young Italian female voice Luca im_nicola Italian Male Young Italian male voice Mei zf_xiaobei Chinese Female Young Chinese female voice Lian zf_xiaoni Chinese Female Young Chinese female voice Wei zm_yunjian Chinese Male Young Chinese male voice Camila pf_dora Portuguese Female Young Portuguese female voice Thiago pm_alex Portuguese Male Young Portuguese male voice And many more voices available! This demo features our most popular voices, with additional options continuously being added. How Kokoro TTS Works: The Technical Breakthrough Kokoro achieves its remarkable efficiency through a revolutionary architectural design that combines the best elements of StyleTTS 2 and iSTFTNet in a decoder-only approach: Innovative Architecture Hybrid Design: Merges StyleTTS 2's transformer-based decoder with iSTFTNet's efficient vocoder for optimal quality-to-size ratio Decoder-Only Architecture: Eliminates the computationally expensive diffusion-based style modeling and separate text encoders that other models require Streamlined Waveform Generation: Uses iSTFTNet for fast and efficient audio synthesis without quality compromise High-Quality Training Data: Trained exclusively on carefully curated, permissive/non-copyrighted audio data focused on long-form narration This innovative approach enables Kokoro to generate 24kHz high-fidelity audio with minimal computational resources, redefining what's possible in open-source text-to-speech technology. Comparison with Traditional TTS Architectures Unlike classical TTS models such as Tacotron 2 (which uses slow attention-based mel generation and requires a separate vocoder) or FastSpeech 2 (which relies on a two-stage pipeline and teacher-forced alignments), Kokoro's streamlined architecture generates speech in one efficient pass. By removing diffusion processes and autoregressive bottlenecks, Kokoro achieves superior speed without sacrificing quality. This makes it uniquely positioned for both real-time applications and batch processing of long-form content. Why Choose Kokoro TTS? ⚡ Unmatched Performance and Efficiency Kokoro TTS delivers remarkable speed that makes it perfect for real-time applications and large-scale content production: Outstanding GPU Performance: ~210× real-time on high-end GPUs (RTX 4090), ~90× real-time on consumer GPUs like the 3090 Ti Impressive CPU Performance: 3-11× real-time on modern CPUs, making it viable even without dedicated graphics hardware Ultra-Low Latency: Synthesizes typical sentences in just 40-70ms on GPU, enabling truly interactive applications Exceptional Throughput: Handles 500+ simultaneous requests with response times around 2 seconds, ideal for high-traffic services 💸 Cost-Effective Alternative to Premium Services Free and Open Alternative to ElevenLabs: Achieve professional-grade voice synthesis without expensive subscription fees No Per-Character Pricing: Generate unlimited audio without worrying about usage-based pricing models Local Processing Option: Run entirely on your own hardware without relying on internet connectivity or cloud services Fully Commercial-Ready: Apache 2.0 license permits unrestricted use in commercial products and services 🎯 Perfect For a Wide Range of Applications Content Creation: Generate professional voiceovers for videos, podcasts, YouTube content, and social media Audiobook Production: Convert ebooks, articles, and long-form content to engaging audio in minutes instead of hours Gaming & VR: Add dynamic voice lines to games and virtual reality experiences with minimal latency Accessibility Tools: Build screen readers and assistive technology that sounds natural and engaging Voice Assistants & Chatbots: Create responsive AI interfaces with human-like speech capabilities E-Learning & Education: Develop engaging learning materials with clear, natural audio narration IVR & Telephony Systems: Improve customer experience with natural-sounding automated phone systems Localization & Dubbing: Translate and voice content across multiple languages efficiently What Users Are Saying About Kokoro TTS Kokoro TTS has garnered enthusiastic praise from developers, content creators, and AI enthusiasts across the community: "This thing is crazy for 82M! I generated a six-hour audiobook from a full book in just four minutes. The consistency is incredible across long texts." "Kokoro is the best open-source TTS I've used... really tiny, so with the right hardware it's really fast. Voice quality rivals commercial services I've paid for." "I've tried Fastspeech, VoiceCraft, Coqui before... these required chunking input into short pieces and post-processing to remove pauses. Kokoro just works on long texts without too many issues." "Voice is pleasant and for long texts it reads in a very stable manner without odd pauses or glitches. It's become my go-to for all my content creation needs." "As someone building accessibility tools, Kokoro has been a game-changer. The voices sound natural enough that my users actually enjoy listening to them, unlike the robotic alternatives." How to Use Kokoro TTS in Your Projects Want to integrate Kokoro TTS into your own applications? You have several flexible options: 1. Unreal Speech API (Fastest & Easiest) For production-ready implementation with minimal setup, use the Unreal Speech API powered by Kokoro TTS. Note: You'll need to sign in to get a free API key, which you'll then find on your dashboard. See the API docs for more info. # Endpoint: /stream # - Convert up to 1,000 characters ASAP # - Synchronous, instant response (0.3s) # - Streams back raw audio data (no timestamps) import requests response = requests.post( 'https://api.v8.unrealspeech.com/stream', headers = { 'Authorization' : 'Bearer YOUR_API_KEY' }, json = { 'Text': '''Your text goes here''', # Up to 1,000 characters 'VoiceId': 'af_bella', # Choose from available voice IDs 'Bitrate': '192k', # 320k, 256k, 192k, ... 'Speed': '0', # -1.0 to 1.0 'Pitch': '1', # 0.5 to 1.5 'Codec': 'libmp3lame', # libmp3lame or pcm_mulaw } ) with open('audio.mp3', 'wb') as f: f.write(response.content) Why Choose Unreal Speech API: 11× cheaper than ElevenLabs Stream audio in just 300ms Request up to 10-hour audio files Includes per-word timestamps Production-ready infrastructure No need to manage your own hardware 2. Python Implementation (Self-Hosted) For those who prefer to run Kokoro locally or in their own infrastructure: from kokoro import pipeline # Create the TTS pipeline audio_generator = pipeline( "This is a demonstration of the Kokoro TTS system, which produces remarkably natural speech from a compact 82 million parameter model.", voice="af_bella", speed=1.0 ) # Process the generated audio for _, _, audio in audio_generator: with open("kokoro_demo.wav", "wb") as f: f.write(audio) 3. Command Line Usage For quick generation from the terminal: kokoro-tts -v af_bella "Hello, this is Kokoro speaking. I'm a compact but powerful text-to-speech system." -o output.wav System Requirements Kokoro TTS is remarkably efficient, making it accessible on a wide range of hardware: CPU: Modern multi-core CPU for real-time speeds (8+ cores recommended for optimal performance) GPU: Even mid-range cards like GTX 1060 (6GB) can handle Kokoro efficiently, with high-end cards achieving 100-200× real-time speeds Memory: ~2GB RAM for model and audio processing (more for handling very long texts) Disk Space: ~350MB for model plus a few MB for voice files Supported Platforms: Windows, macOS, Linux, and cloud environments Kokoro TTS vs. Other Text-to-Speech Models See how Kokoro compares to other popular TTS solutions: Feature Kokoro TTS XTTS v2 MetaVoice ElevenLabs Model Size 82M parameters 467M parameters 1.2B parameters Proprietary Speed Up to 210× real-time ~30× real-time ~20× real-time Cloud-based Local Deployment ✅ Yes ✅ Yes ✅ Yes ❌ No Quality Ranking 1st in TTS Spaces Arena Lower ranked Lower ranked High quality Commercial License ✅ Apache 2.0 ❌ Restricted ❌ Restricted ❌ Paid service Voice Cloning ❌ No (without fine-tuning) ✅ Yes ✅ Yes ✅ Yes Cost Free (open-source) Free Free Subscription-based Long-form Handling ✅ Excellent ⚠️ Requires chunking ⚠️ Variable quality ✅ Good Resource Usage ✅ Very low ⚠️ Moderate ❌ High N/A (cloud) Multilingual ✅ 8+ languages ✅ Multiple ✅ Multiple ✅ 29+ languages Technical Deep Dive: What Makes Kokoro Special For those interested in the technical details, Kokoro's success stems from several key innovations: Efficient Architecture Kokoro's hybrid model combines the strengths of StyleTTS 2 and iSTFTNet while eliminating their inefficiencies. By removing the diffusion-based style modeling of StyleTTS2 and using iSTFTNet for efficient waveform generation, Kokoro dramatically reduces complexity while preserving quality. Unlike traditional two-stage TTS pipelines (text-to-spec followed by vocoding), Kokoro streamlines the process with a transformer-based decoder that can directly produce audio features with integrated vocoding. This avoids Tacotron's alignment issues and slow iterative output. Benchmarks & Performance Kokoro has proven its merit in head-to-head evaluations, achieving 1st place in the HuggingFace TTS Spaces Arena for single-speaker speech quality. Listeners consistently ranked Kokoro's output above much larger models in blind tests. In Elo-style comparisons of naturalness, Kokoro-82M emerged as a top model, even beating systems trained on vastly more data. For example, "Fish Speech" (trained on ~1 million hours) failed to match Kokoro's naturalness, despite Kokoro being trained on <100 hours of curated data. Training Efficiency Kokoro's training process was remarkably cost-effective, requiring only ~500 GPU hours on A100 hardware (approximately $400). This efficiency demonstrates that with the right architecture and high-quality data, smaller models can achieve state-of-the-art results. Limitations and Future Improvements While Kokoro TTS is impressive, we believe in transparency about its current limitations: Limited Expressiveness: Speech can sound somewhat neutral in emotional range compared to professional voice actors No Built-in Voice Cloning: Cannot mimic new voices without fine-tuning (unlike some commercial options) Multilingual Quality Variations: While supporting multiple languages, quality may vary across non-English languages Short Input Quirks: Performs best with longer texts rather than single words or very short phrases The Kokoro community is actively working on addressing these limitations in future updates, with plans for more expressive models and improved voice variety. Get Started with Kokoro TTS Today Try our live demo above and experience the future of open-source text-to-speech technology. With Kokoro TTS, you can generate professional-quality voiceovers, create accessible content, and build voice-enabled applications without breaking the bank. Ready for Production Use? For production-ready API access with enterprise reliability, ultra-fast response times, and cost-effective pricing, check out Unreal Speech - the premium Kokoro-powered TTS API that's: 11× cheaper than ElevenLabs Streams audio in just 300ms Supports requests up to 10 hours long Includes precise per-word timestamps Backed by enterprise-grade infrastructure Get Started for Free → Frequently Asked Questions About Kokoro TTS What makes Kokoro TTS different from other text-to-speech services? Kokoro TTS stands out for its remarkable efficiency—achieving professional-quality speech with just 82 million parameters (compared to models 5-15× larger). This lightweight design enables fast processing through our API while still outperforming much larger models in quality benchmarks. Our online demo lets you experience Kokoro's capabilities instantly and download the generated MP3s. Unlike most commercial services, the underlying Kokoro model is open-source under the Apache 2.0 license, while our Unreal Speech API provides a production-ready implementation with affordable pricing. Which languages and voices does Kokoro TTS support? Kokoro TTS currently offers 48 voices across 8 languages. You can generate speech in American English, British English, French, Hindi, Spanish, Japanese, Chinese, and Portuguese. Each language includes multiple male and female voices with different characteristics and speaking styles. The voice selection is constantly expanding, with regular updates adding new options and improving existing ones. Can I download and use the generated speech files for my projects? Yes, all audio generated by Kokoro TTS Studio can be freely downloaded as MP3 files and used in both personal and commercial projects. You can use these audio files for YouTube videos, podcasts, e-learning content, audiobooks, or any other application. The following terms apply, based on your subscription plan: Free plan – You must attribute Unreal Speech by including a link to "unrealspeech.com" in the description. Paid plan – You do not need to include any attribution. How do I get the best quality results from Kokoro TTS? For optimal results with Kokoro TTS, use longer sentences or paragraphs rather than single words (the model performs better with context). Include proper punctuation to help with natural pausing and intonation. Experiment with different voices—some may pronounce certain words or phrases more naturally than others depending on your text. For professional applications requiring even higher quality or custom voices, consider Unreal Speech's API which builds upon Kokoro's technology with enterprise-grade reliability. Can I run Kokoro TTS offline on my own computer? Yes, Kokoro TTS can be installed and run locally on your computer without an internet connection. The model is small enough (about 350MB) to run efficiently on most modern computers, even without a dedicated GPU. For local installation, you can use the Python implementation (pip install kokoro) or command-line tools. This makes Kokoro ideal for privacy-conscious users, offline applications, or scenarios where consistent generation without reliance on external services is important. Unreal Speech Studio Powered by Unreal Speech Help Center Contact Us Privacy Policy Terms of Service Sign In Sign in to enable these extra features: Download audio files 250,000 free characters Upgrade to Pro Get these features when you upgrade: Download MP3 (high quality) 1 million free characters 30,000 chars per voiceover Upgrade to Pro 3-day free trial, then $10/monthCancel anytime Download Audio By downloading, you agree to attribute Unreal Speech by including a link to unrealspeech.com in the description of any published content. Download --- API Docs Pricing Studio Blog Sign In Dashboard PersonalDeveloper Current Plan Free Plan and Billings Upgrade Remaining Number of Characters • • • Got a question, or need help? High Volume Inquiry → Go through the onboarding again: Start Onboarding → We value your feedback! Share your experience with our service on G2.com and earn a reward of 3 million characters! Contact us with your review confirmation to claim your reward. Write a Review → Unreal Speech Studio Make studio-quality voice overs for podcasts, videos & more. Open Studio → API Key Check out the docs for API info and code samples: API Docs → Made with ❤️ in SF Powered by Kokoro TTS API Docs Pricing Studio Contact Blog Compare Browse AI Apps Affiliate Recent Posts Integrating FastSpeech 2 for Text-to-Speech Synthesis with Fairseq and Hugging Face Exploring the Potential of GPT-SoVITS-Fork for Text-to-Speech Applications Exploring the GPT-SoVITS Kancolle Zuikaku TTS Model: A Comprehensive Guide Exploring Voice Synthesis with ESPnet: A Deep Dive into the kan-bayashi_csmsc_fastspeech Model Introducing OpenVoice: Revolutionizing Text-to-Speech with Instant Voice Cloning and Multilingual Capabilities How to Leverage Twelve Labs API for Effortless YouTube Video Summaries, Chapters, and Highlights API Docs Pricing Studio Blog Sign In Sign In Copied to clipboard!