IndexTTS 1
by Bilibili
Free to use, every day. A few seconds of audio recreates a voice faithfully — even from a short clip. Best for bulk dubbing and long-form narration.
BiliVoice: Free AI Voice Clone
- Free daily credits on every model — clone a voice and dub your first script today
- 5 of the top 10 speech models — IndexTTS2.5, OmniVoice, VoxCPM2 and more, all behind one API
- Our own H100 cluster — RTF under 0.2, and 120K characters for $0.0014
5 Models5 of the Top 10
2B characters processed daily
120K chars per $0.0014
60K hours of audio daily
Model Count5 of the Top 10
Generation ScaleCost spread by scale · cheaper and more stable
Compute PowerIn-house H100 · RTF<0.2
Leading on Three FrontsModels · Scale · Compute
Every Model in One Website
Bilivoice is an aggregation platform, not a single model: five of the ten most popular speech models sit behind one API, so you can compare them side by side, switch between them with one click and generate without leaving the page. The badge on every card shows where the model ranks on the TTS leaderboard at voicebox.xin, and who built it. Every model runs on free daily credits.
Bilivoice Aggregation Engine
TTS Aggregation · Unified API
Every model is called through one platform with a unified parameter format, unified audio output and unified pricing. Native support for mainstream agent platforms including OpenClaw, Coze and WorkBuddy.
- Unified API
- Native Agent Support
- One-Click Model Switch
- Unified Billing
IndexTTS2
by Bilibili
Emotion you can dial in: pick a tag, set the intensity, keep the voice you cloned. Best for game dialogue, short dramas and character-driven audiobooks.
IndexTTS2.5
by Bilibili
The latest IndexTTS release — stronger cloning, finer emotion, better quality, no tuning. Best as your default for everyday dubbing and narration.
OmniVoice
by Xiaomi
600+ languages and 20+ dialects, with a dialect mode most models lack, plus multi-character dialogue. Best for localised dubbing, regional video and large-cast projects.
VoxCPM2
by OpenBMB
High-fidelity audio at high bitrates, across 30 languages, fast enough for long batches. Best for podcasts, ads and tracks that get mastered.
More Models on the Way
We keep close track of what is happening at the frontier
Hear Real Voices
Listen to high-quality voice samples generated by Bilivoice.
Highly Realistic Voice Cloning
Original Audio
Cloned Audio
ClonedHow to achieve highly realistic voice cloning results?
Use a supported Bilivoice model such as IndexTTS2.5 or OmniVoice to achieve highly realistic voice cloning results.
Cross-lingual Cloning
Original Audio
Spanish
MultilingualChinese
MultilingualJapanese
MultilingualKorean
MultilingualFrench
MultilingualGerman
MultilingualHow to enable cross-lingual voice cloning?
Use a supported multilingual Bilivoice model to enable cross-lingual voice cloning. Language coverage varies by model.
Emotion & Style Control
Original Audio
Happy
EmotionCalm
EmotionSad
EmotionAngry
EmotionMelancholic
EmotionDisgust
EmotionFearful
EmotionSurprised
EmotionHow to add emotions and control voice style?
Use IndexTTS2.5 emotion controls to fine-tune voice mood, intensity and delivery style.
Accent Transformation
Original Audio
British English
AccentAustralian English
AccentChinese English
AccentIndian English
AccentRussian English
AccentCanadian English
AccentJapanese English
AccentAmerican English
AccentPortuguese English
AccentKorean English
AccentHow to use different accents with the same voice?
Use a supported multilingual Bilivoice model to preserve the voice while switching languages and accents.
Demo samples are AI-generated for showcasing Bilivoice voice cloning only, and do not represent any real person or brand endorsement.
In-House H100 Cluster · Leading Compute
We run our own NVIDIA H100 cluster, with full control of the stack from bare metal to inference framework. Most models reach RTF below 0.2 — 10 seconds of audio in 2 — which removes relay latency, throttling and instability at the source.
In-House H100 Cluster
Our own NVIDIA H100 cluster runs fully loaded around the clock. No third-party throttling, no queueing, no service degradation — just as smooth at peak hours.
Zero Relay · Instant Response
Requests go straight to our own compute nodes: no proxy layer, no API forwarding chain. Most models sit below RTF 0.2, so 10 seconds of audio takes 2 and long text is ready in seconds.
Ultra-High Stability
Hot standby across datacenters plus smart load balancing holds 99.9% availability while handling 2B characters a day. If a node fails, users are moved over without noticing.
Ultra-Low Cost
We cut out the reseller margin and pass the compute savings straight on. $0.0014 buys 120K characters, so 10,000 characters costs under $0.14 — well within reach.
2B UTF-8 characters handled daily
99.9% Service Availability
<1s Average Response Latency
RTF<0.210s audio · 2s to generate
Bilivoice vs ElevenLabs
The same 120K characters cost $0.0014 on Bilivoice and about $12 on ElevenLabs. Eight of the top ten open models sit behind one API, all served from our own H100 cluster.
| Feature | Bilivoice Best value | ElevenLabs |
|---|---|---|
| Cost of 120K characters | $0.0014 | ≈ $12.00 |
| Price per 1M characters | $0.012 | ≈ $100 |
| TTS models available | 5 of the top 10, switchable | 1 proprietary family |
| Voice cloning | 20-second sample, unlimited voices | Instant clone; Pro clone from Creator |
| Languages | 600+ languages & 20+ dialects | 29 languages |
| Open-source models | Yes — IndexTTS, OmniVoice, VoxCPM2… | No |
| Compute | Own H100 cluster, zero relay | Managed cloud |
| Free tier | Free daily credits | 10K chars/month, non-commercial |
| Unified API | One API, unified billing | One API, per-model rates |
ElevenLabs figures are taken from its published TTS API rate of $0.10 per 1,000 characters and its public plan quotas (September 2026). Bilivoice figures use the same 120K-character baseline sold for $0.0014.
Why Bilivoice
Ahead on all three — model count, generation scale and compute — supporting your voice end to end
Model Count
Five popular models including IndexTTS 1, IndexTTS2, IndexTTS2.5, OmniVoice and VoxCPM2 — 5 of the top 10. Most of the popular ones are here, and you can switch freely on one platform instead of comparing across tools.
Generation Scale
We handle 2B UTF-8 characters and 60K hours of audio every day. That scale is what makes $0.0014 per 120K characters possible, and that volume of real traffic is what keeps stability and availability at 99.9% — batch jobs of any size hold up.
Compute Power
Our own NVIDIA H100 cluster with zero-relay direct inference. Most models reach RTF below 0.2 — 10 seconds of audio in 2 — so even batch work finishes without a long wait.
A generational dubbing tool that millions rely on
Creators across industries came for the same three leads: model count, generation scale and compute
I can audition one script across 5 models — IndexTTS2, IndexTTS2.5, OmniVoice and the rest — and switch whenever the timbre isn't right. Most of the popular models are here, so changing dubbing styles for short videos is instant and my output has doubled.
The H100 cluster really is fast: most models below RTF 0.2, 10 seconds of audio in 2. Solo narration is ready the moment I upload, so my release schedule never waits on dubbing — and listeners say the immersion is way up.
An audiobook runs to hundreds of thousands of characters and I was braced for expensive and flaky. Bilivoice's 2B-characters-a-day scale spreads the cost so thin that 10,000 characters runs under $0.14, and 60K hours of daily output is proof enough of the stability. Long batches finish in one pass — no more worrying.
Our whole corporate course library — hundreds of hours — is produced on Bilivoice. Self-operated compute with zero relay keeps the API fast and steady, the voice stays consistent and delivery is on time. The training-video pipeline is one thing I no longer worry about.
Game dubbing needs a lot of distinct characters, and five aggregated models give me plenty of voices to pick from — nearly every name on the popularity chart is here. Real-time tuning in the browser keeps it cheap and effective, and it has sped up development considerably.
My daily uploads run entirely on it: the popular models are basically all here, so I pick one and stay with it. Sub-0.2 RTF makes the dubbing step feel instantaneous. Posting daily went from pressure to pleasure, and the tone of the show stays consistent.
Questions, Answered
What you probably want to know about Bilivoice
Bilivoice is a leading voice cloning and dubbing platform. Where single-model tools give you one option, Bilivoice aggregates the most popular speech models — 5 of the top 10 — and runs all its own compute, handling 2B UTF-8 characters a day. Scale spreads the cost and volume proves the reliability: cheaper than comparable tools, with none of the latency or instability that comes from relaying someone else's API. One platform, one API, one bill, every model.
Bilivoice handles bilingual dubbing natively and also covers a range of dialects. Cross-lingual conversion keeps your voice's character fully intact when you switch languages.
Three steps: 1. Record about 20 seconds of clean speech with no noise and no background music. 2. Upload it on the Create Voice page and give it a name. 3. Select your voice on the dubbing page, type your text, and generate.
Yes. The models are trained specifically to work out the right reading of most common heteronyms from context, so you never have to annotate pronunciation by hand.
Around 20 seconds is ideal. The audio should have no ambient noise, no background music, no other speakers and no reverb. Better samples give better results — if yours falls short, run voice separation or denoising first.
Three days for files generated in the console, one day for files generated through the API — download them promptly. Voice samples and voice models are kept for the life of your membership, and for one month after it ends.
Yes. Bilivoice exposes a full API so you can build voice cloning, dubbing and emotional speech synthesis into your own product. Because we run all our own compute and most models sit below RTF 0.2, response times and stability are far ahead of relay-based alternatives. Talk to us about pricing and onboarding.
Voice Cloning & AI Dubbing, Ahead of the Field
Bilivoice is a leading voice cloning and dubbing platform, ahead on three fronts: model count, generation scale and compute. It aggregates five popular models — IndexTTS 1, IndexTTS2, IndexTTS2.5, OmniVoice and VoxCPM2, 5 of the top 10 — while scale spreads unit cost and volume proves reliability (2B UTF-8 characters and 60K hours of audio a day, close to seven years of continuous playback), which is why it is both cheaper and steadier. We run our own NVIDIA H100 cluster with no API relay and hold 99.9% availability. An international team of leading AI scientists, speech specialists and senior product engineers, drawing on deep work in multimodal models and speech synthesis, has pushed past the bottlenecks of low-resource high-fidelity dubbing to give enterprises and creators worldwide one fast, lifelike and dependable place to do AI voice.
Model Count Leads · Every Model in One Place
Since 2016 we have brought together engineers from Google, Microsoft and Tsinghua to build an original speech architecture. Five popular models — IndexTTS 1, IndexTTS2, IndexTTS2.5, OmniVoice and VoxCPM2, 5 of the top 10 — reproduce a voice from three seconds of audio, capturing emotional prosody and the detail of breath.
Compute Leads · In-House H100 Cluster
Our own NVIDIA H100 cluster with hot standby across datacenters and smart load balancing. Most models reach RTF below 0.2 — 10 seconds of audio in 2 — at 99.9% availability, with relay latency designed out from the start.
Scale Leads · Cheaper and Steadier
Across the product line we handle 2B UTF-8 characters and synthesize 60K hours of audio daily — close to seven years of continuous playback. Scale spreads the unit cost, so professional dubbing is priced for everyone, and real traffic validating the system day after day keeps stability and availability where they should be.