Fish Audio Reviews & Product Details

What I like most about Fish Audio is how it balances a clean, no-frills web UI (playground, Story Studio for multi-character long-form narration, and a 2M+ community voice library) with genuinely production-grade engineering: the S2/S2.1-Pro models deliver expressive, emotion-tagged TTS and 10–15s zero-shot voice cloning that beats ElevenLabs on prosody in several non-English languages, while the REST + WebSocket API with official Python/TypeScript SDKs, LiveKit/Pipecat hooks, llms.txt/OpenAPI specs and an agent skill for Cursor/Claude/Codex make it one of the most developer-friendly voice stacks out there, all running at sub-500ms streaming latency suitable for real-time voice agents and conversational AI. Performance-wise it scales from free prototyping to pay-as-you-go API at ~$15 per 1M UTF-8 bytes and Plus/Pro tiers from ~$11–$100/month, which is roughly a sixth of ElevenLabs’ cost for comparable output, and the open-weight Fish-Speech/S1/S2 models let teams self-host in VPC or air-gapped environments when they need data residency. After-sales support is lighter than enterprise incumbents—email plus docs, no white-glove SLA on lower tiers—but the open-source heritage (So-VITS-SVC, Bert-VITS2 lineage), transparent blind-test benchmarks (Audio Turing score 0.515 on S2.1 Pro) and active Discord/Hacker News/Reddit community partly compensate for the lack of hand-holding. On agents specifically, it’s a standout: inline direction tags like [whisper] or [chuckle] travel inside the text, ASR+TTS+clone live in one stack, and interruption-aware streaming makes it easy to wire into Retell/HeyGen-style assistants, so the thing I’d keep over any competitor is the combination of low-cost expressive quality, open weights, and an API surface that treats voice as a programmable performance layer rather than a black-box narration button. Review collected by and hosted on G2.com.

What I like least about Fish Audio is that the things that make it appealing to developers—open weights, byte-metered API, inline emotion tags, concurrency-gated tiers—are exactly what make it feel rough around the edges for everyone else: the web UI is functional but unpolished compared to ElevenLabs’ Studio, with no real waveform editor, light post-processing, or non-technical project-management layer, so creators still bounce to external DAWs; the 2M+ community voice library is huge but unmoderated, so commercial users have to manually audition voices and carry the licensing/consent risk themselves since public voices are not automatically cleared; integrations cover REST/WebSocket/Python/TS plus LiveKit/Pipecat/MCP hooks but stop short of first-class LangChain/LlamaIndex wrappers, deeper no-code connectors, or the bundled STT-LLM-TTS orchestration that more mature voice-agent stacks ship, leaving you to wire interrupt handling, retry, and concurrency math yourself against a prepaid-spend concurrency ladder rather than simple QPS; performance is genuinely fast (sub-150ms TTFA) yet English long-form stability and smaller-language fidelity still trail ElevenLabs, emotion tags occasionally fire inconsistently, and 15-second clones degrade sharply on noisy reference audio; pricing is cheap on paper (~$15/M UTF-8 bytes) but the credit/byte/audio-hour mixing, non-rolling monthly quotas, and free tier capped at a few thousand characters per day make real cost modeling annoying for irregular teams; after-sales is email-plus-docs with no published SLA on self-serve tiers, and the compliance story (no clear SOC 2/HIPAA/ISO posture, thin DPA, ambiguous prompt-training opt-out) blocks healthcare/finance/government buyers despite the self-host escape hatch; and on agents specifically, while inline tags and streaming latency are great for prototyping, production voice agents still need you to build turn-taking, barge-in, session memory, and fallback logic on top, with S2’s removed LoRA fine-tune path and tag-placement sensitivity meaning you trade ElevenLabs-style “set stability/clarity/style sliders and forget it” for more prompt engineering and more eval work—so the honest downside is that Fish Audio is a fantastic programmable voice engine, not a finished enterprise narration or conversational-platform product. Review collected by and hosted on G2.com.

The voices just sound less fake than the other TTS tools I've tried, including ElevenLabs. I can still tell if I listen for it, but it's close enough that I stopped re-recording every line. I use it for short explainer and demo voiceovers. The script changes a lot, so I was spending more time on the mic than on the actual video. Cloning from a short clip is pretty convenient too, the voice stays consistent across videos. I mess with the emotion tags when something comes out too flat, though those don't always land on the first try. Review collected by and hosted on G2.com.

I end up regenerating a lot more than I thought I would. A line sounds fine except one word comes out weird, then the next take messes up a different word. When I'm trying to finish a video the same night, that gets old fast. I still mess with the emotion tags but I'm kind of guessing half the time. I wish I could keep the parts that already sound good and only reroll the bad word. Review collected by and hosted on G2.com.

The voice sounds natural, and it’s much, much cheaper than the leader in AI voice generation. The UI is pretty intuitive, especially if you've used Elevenlabs. The API works well, I've used it in my video production pipeline. Support can be better, it's currently using discord but lots of free users asking for credits. The generation is pretty good, with emotions and stuff, although quality can be more consistent. Review collected by and hosted on G2.com.

There’s still a lot of gibberish and quite a few bugs with the API. Also, the voices tend to end with an uptalk tone. With voice cloning, I’m also hearing voices that sound out of tune at times, for some reason. I don't like their support system, which uses Discord, and messages get lost in a sea of messages. The pricing tier can be better, there should be a mid-tier range. Review collected by and hosted on G2.com.

What I like most is the exceptional quality and naturalness of the voice cloning. With only a few seconds of reference audio, it accurately captures subtle emotions, timbre, and a natural speech rhythm, so the generated voices sound convincingly human. Review collected by and hosted on G2.com.

Occasionally, the model produces audio hallucinations, unexpected skips, or repetitive artifacts, particularly when synthesizing longer paragraphs. Better consistency across extended passages would make it feel much more reliable and dependable overall. Review collected by and hosted on G2.com.

What stands out most about Fish Audio is the accuracy and natural intonation of its voice cloning. Even with a short reference audio sample, it captures nuanced emotions, pacing, and speech quirks very convincingly. The API response time is fast, and the web interface is clean and straightforward to use. Multilingual voice synthesis handles switching between accents and languages much smoother than many traditional TTS engines. Review collected by and hosted on G2.com.

When generating longer paragraphs, the tone can occasionally drift or produce slight background artifacts, requiring manual slicing of text into smaller segments for consistent quality. The documentation for advanced API parameters and edge-case error handling could also be more comprehensive for developers. Review collected by and hosted on G2.com.

I’ve been using fish-audio’s product on my 4070 Ti Super graphics card ever since it was first open-sourced. Over the past two months, they’ve been offering free API credits, which has made them the only voice provider for my Hermes agent. When the free period ends, I’m willing to pay for the service.

We all know that current LLM responses aren’t always perfect, but adding specific timbres makes them sound completely different. It gives the whole system a sense of soul and personality. Review collected by and hosted on G2.com.

I honestly think it would be even better if they released more open-source ONNX inference models for CPU use. That would really encourage more people to start using TTS models, and once they try them, they likely won’t want to go back. Review collected by and hosted on G2.com.

The audio tags are awesome; the way you can enhance the results by using these tags just makes Fish Audio way better than any other model I've used in the past. Review collected by and hosted on G2.com.

It's not a total dislike, but not really knowing if voices in the catalog fail to comply with copyright laws might be the only concern I have Review collected by and hosted on G2.com.

What I like best is honestly the voice realism ,it makes a real difference when you're narrating scripture, since anything that sounds robotic breaks the meditative tone immediately. The Python SDK is genuinely easy to work with too; I had it integrated into my pipeline within an afternoon, and it's saved me a ton of manual recording time.

Speed matters a lot for how I use it. My pipeline runs mostly unattended, generating audio in concurrent batches, and Fish Audio holds up well under that load without the latency becoming a bottleneck. Getting started was refreshingly simple API key, voice ID, and I was already generating usable audio.

We actually switched over from ElevenLabs mainly because of cost. For a pipeline that's running daily and generating a high volume of audio, Fish Audio's pricing made a lot more sense without giving up quality Review collected by and hosted on G2.com.

Pronunciation of certain English terms can be hit or miss sometimes. Numbers are a specific pain point , for example, "7000" sometimes gets read out as "seven zero zero zero" instead of "seven thousand," and the behavior seems inconsistent depending on formatting, whether it's written as 7000, 7,000, or attached to something like a year or unit.

The bigger issue for me is that when I'm running the pipeline fully hands-off with concurrent requests, a handful of segments occasionally come out sounding noticeably robotic , off enough that I have to catch them, delete them, and manually restart that part of the pipeline. For a setup that's supposed to run unattended, that manual step defeats the purpose a bit.

What would actually fix this for me is some kind of confidence score per segment, so I could programmatically tell whether the output is genuinely close to the voice clone or not. Even better would be if Fish Audio handled retries server-side automatically when a segment falls below quality, instead of leaving it to me to catch on the client side. Review collected by and hosted on G2.com.

I appreciate that Fish Audio generates my voice overs well, observing punctuations and story flow, especially with the S2 model. I like the smoothness and speed of generation, the quality of the audio, the very cheap pricing, and the ease of use for both the platform and API integration. It saves me time and money, making me worry less. The initial setup was really smooth without any confusion. Review collected by and hosted on G2.com.

One thing I think needs to be improved are the number of available default voices, it will really help very much. I need more default options more specifically storytelling voices. Different varieties, voices that can capture my audience when they watch my videos would be great. I've seen such voices in other people's videos; they are really calm and make you feel like you are sinking into the story as you hear them, not rushed. Review collected by and hosted on G2.com.

What I like best about Fish Audio is the natural quality of the generated voices. The speech sounds expressive and human, with realistic pacing, pronunciation, and tone, which makes it suitable for professional video narration.

I also value the voice cloning feature because it allows me to maintain a consistent voice across different projects without recording every script manually. This has significantly improved my workflow, saving me many hours when producing long-form videos and short content.

The platform is easy to use, generates audio quickly, and offers a strong balance between quality and cost. For me, the biggest benefit is being able to create professional and consistent voiceovers much faster than with a traditional recording process. Review collected by and hosted on G2.com.

Occasionally, pronunciation and emotional consistency can vary in longer scripts. More control over pauses, pacing, and emphasis would make the editing process even better. Review collected by and hosted on G2.com.