Head to head · Audio · updated Oct 26, 2026
Fish Speech vs Deepgram
Fish Speech scores 68 to Deepgram's 15 on public evidence for text-to-speech & voice cloning, and they rank #1 and #4 there. Both are free to start. Numbers below are the same public signals the ranking uses, recomputed nightly — nothing here is sponsored.
Fish Speech
Open multilingual text-to-speech with few-shot voice cloning.
+ Clones a voice from seconds of audio and runs on your own GPU.
− The licence restricts some commercial use and setup is research-grade.
Deepgram
A speech-to-text and text-to-speech API built for low-latency streaming at scale.
+ Among the fastest real-time transcription APIs available, with diarization and keyword boosting built in.
− Developer-only; there is no end-user app, and accuracy on heavily accented speech trails Whisper-class models.
Side by side
public evidence| Fish Speech | Deepgram | |
|---|---|---|
| Evidence score / 100 | 68 | 15 |
| GitHub stars | 33,412 | closed source |
| Hacker News stories, 30d | 0 | 1 |
| Release-feed mentions, 30d | 0 | 0 |
| Last activity | 12 days ago | – |
| Pricing | Open source · Weights are free; fish.audio hosts it for a fee | Free tier · $200 free credit; Nova STT from about $0.0043/min |
| Platforms | cli, linux, api, web | api |
| Maker | Fish Audio | Deepgram |
Which one?
Pick Fish Speech if
- you want the stronger public evidence (68 vs 15)
- you want open source — self-host it, read it, fork it
- you need it on cli, linux, web
- a bigger GitHub community matters (33,412 vs 0 stars)
Best for self-hosted voice cloning.
Pick Deepgram if
Among the fastest real-time transcription APIs available, with diarization and keyword boosting built in.
Best for real-time voice apps.