Fish Audio: realistic speech without the studio bill

Estimated reading time: 11 minutes

Key takeaways

• Fish Audio covers text-to-speech, voice cloning, multilingual speech, speech-to-text, audio separation, and sound effects in one platform.
• You can use it in a web app, mobile app, REST API, or SDK. Creators and developers both have a path in.
• Models include S1, S2, and S2-Pro (open-weight Fish Speech family). You can use the hosted product or self-host if your team has the infra.
• Voice cloning is fast: some flows work from ~10–15 seconds of sample audio, with output that holds up in blind tests against pricier tools.
• Emotion tags, pacing, pauses, and word-level control let you shape delivery without re-recording.
• Common buyers: creators, businesses, game devs, and support teams building voice agents.
• Independent comparisons often rank Fish Audio high on quality, latency, and price per character relative to ElevenLabs-class alternatives.

Table of contents

1. What Fish Audio is
2. Core capabilities
3. Models and technology
4. Voices, languages, and scale
5. Platform, apps, and developer workflows
6. Pricing and plans
7. Typical use cases
8. Independent reviews
9. Company background
10. Ethical and legal considerations
11. Related tools on 101aitools
12. FAQ

---

Fish Audio turns text (or a short voice sample) into speech and audio assets you can ship. The pitch is simple: AI voices that sound human enough for YouTube, games, and customer support, without booking studio time.

What it actually does:

• Text-to-speech (TTS)
• Voice cloning
• Multilingual speech and dubbing
• Speech-to-text (ASR)
• Audio separation and sound effects

Browser, mobile app, REST API, SDK. No-code creators and engineers can both get value from the same stack. See the full listing on 101aitools for pricing and category context alongside other voice tools.

---

What Fish Audio is

At the core, Fish Audio converts text or a voice sample into natural speech with minimal setup. The usual buyers: creators, marketers, editors, product teams, developers, and ops teams rolling out voice features.

Three ways people use it:

• Web app: paste text, pick a voice, generate, download
• REST API and SDKs: embed TTS, streaming, ASR, and cloning in your product
• Mobile app: same workflow on iPhone/iPad when you're away from a desk

Why teams pick it over a single-purpose TTS API:

• Multiple audio tools in one place (TTS, cloning, ASR, separation, effects)
• Built on open-weight Fish Speech models, with a commercial layer at fish.audio
• Fits narration, dubbing, game dialogue, and live voice agents

Useful starting points:

• Capabilities overview
• Voice cloning
• Developers hub

---

Core capabilities of Fish Audio

This is not "type text, get MP3." The product is built around expressive speech, fast cloning, and enough control that you can fix a line without regenerating the whole clip.

Text-to-speech

High-fidelity output for voiceovers, ads, podcasts, audiobooks, and game dialogue. Generation is fast, and streaming works for near real-time use cases (support bots, interactive apps).

Voice cloning

Synthetic voices from short samples. Some models work from ~10 seconds; the marketing site cites ~15 seconds for instant clone. Useful when you need one brand voice, one game character, or one narrator across languages.

Emotion and style control

Inline tags adjust tone (calm, excited, whisper, laugh), pacing, and pauses. S2 adds word-level control: tag a single word or phrase to change prosody without touching the rest of the sentence. That matters when a script has one emphasized word or a dramatic pause mid-line. Noiz AI takes a similar emoji-tag approach if you want a second opinion on emotion-driven TTS.

Multilingual speech

Broad language coverage (often cited as 80+ languages). Cross-lingual cloning keeps the same voice identity when you localize: clone in English, generate in Japanese or Spanish with the same "speaker."

Other audio tools

ASR, audio separation (vocals vs instruments), voice changing, and sound-effect generation. Handy if your pipeline is "transcribe, separate, regenerate" rather than TTS-only.

Further reading:

• S2 word-level control
• Capabilities docs

---

Models and technology behind Fish Audio

Fish Audio runs on the Fish Speech model family: S1, S2, and S2-Pro.

S1

Early commercial model. Established the baseline for expressive TTS and emotional delivery.

S2 / S2-Pro

S2 improved realism, multilingual quality, cloning, and control. S2-Pro is the higher tier: more stable long-form output and stronger expressiveness. S2.1 Pro is the current flagship on the consumer site.

Word-level control

Tag individual words or phrases to change prosody, pause, speed, or emotion. Less "regenerate until it sounds right," more surgical edits.

Open weights and self-hosting

Teams with GPU infra can self-host available weights for privacy, cost at scale, or custom pipelines. Most teams still use the hosted API because billing and ops are simpler.

Latency and streaming

Built for low-latency streaming. Useful for live assistants and anything where time-to-first-audio matters. Sub-second startup is commonly reported; your mileage depends on text length, model, and region.

Docs: Models overview and pricing

---

Voices, languages, and scale in Fish Audio

Voice scale

The platform hosts a large community voice library (2M+ voices cited on site). Directories sometimes inflate that into "millions of configurations." In practice: lots of preset voices plus cloning, so you're not stuck with the same 20 stock narrators.

Languages

English, Japanese, Korean, Chinese, French, German, Arabic, Spanish, and many others. Exact count varies by source (30+ on API comparisons, 80+ on marketing). Check the model docs for your target locale before you commit.

Cross-lingual cloning

Clone once, generate in another language. Important for localization teams that want one character voice across regions without re-hiring voice talent per market.

Why scale matters

Faster iteration, consistent brand or character voice, and less "we need a new recording session for every script change."

References:

• Voice cloning
• Tool directory listing

---

Platform, apps, and developer workflows

Web app

Paste text, choose voice, add emotion/pacing tags, generate, download. Straightforward for creators who don't want to touch an API.

Mobile app

iOS app supports TTS, cloning, and multi-speaker scripts.

Developer tools

REST API plus SDKs (Python, TypeScript/Node.js). Embed TTS, streaming, ASR, and cloning in apps, games, and internal tools.

Typical engineering use cases:

• Streaming TTS in a consumer app
• ASR + TTS pipelines
• Multilingual dubbing automation
• Voice agents and game dialogue systems

Docs: Developers

---

Pricing and plans for Fish Audio

Tiers change; confirm at fish.audio/plan before you budget.

Consumer / creator plans (July 2026)

Plan	Price	What you get (high level)
Free	$0/mo	~8,000 credits (~7 min/mo), 500 chars/generation, personal use per FAQ
Plus	$15/mo ($5.50/mo annual)	~250K credits (~200 min), 1 professional clone slot, commercial use
Pro	$100/mo ($37.50/mo annual)	~2M credits (~1,620 min), 3 team seats, 5 pro clone slots
Max	$999/mo ($749/mo annual)	~25M credits, 10 seats, 15 pro clone slots

Credits reset monthly. Unused minutes don't roll over.

API pricing (pay-as-you-go)

• TTS (S2-Pro, S2.1-Pro, S1): $15 per million UTF-8 bytes (~180K English words, ~12 hours of speech)
• ASR (transcribe-1): $0.36 per audio hour
• Voice design: $0.01 per successful request

That API rate is often 5–10x cheaper per character than premium competitors like ElevenLabs, which is why dev teams evaluate Fish Audio early. For a lower-commitment consumer option in the same price range, see Play.ht or Speechify.

Self-hosting

Open weights can cut API spend or keep data on-prem. You pay in engineering time and GPU cost instead.

Questions before you buy

• How many minutes per month?
• Do you need commercial rights? (Free tier is personal-only per FAQ.)
• API only, web only, or both?
• How many clone slots do you need?

Reference: Models and pricing docs

---

Typical use cases and target users

Creators

YouTube voiceovers, short-form video, podcast clips, narration, character voices. Good when you revise scripts often and don't want to re-record every time.

Audiobooks

Long-form narration with consistent voice and fast revisions. S2-Pro is aimed at hours-long output; still test a chapter before you commit to a full book.

Game development

NPC dialogue, character voices, interactive speech, localization. Emotion tags help when one line needs to land differently than the rest of the scene.

Businesses

Ads, explainers, support bots, training, voice agents. API pricing makes high-volume use cases more viable than with some subscription-only TTS vendors like Murf AI.

Accessibility and localization

Translated audio, screen-reader-friendly content, local-language versions with the same cloned voice.

Why teams switch from studio-only workflows: faster turnaround, reusable voice identity, lower marginal cost per minute. Weaker when you need a specific human celebrity voice with legal clearance, or broadcast union requirements.

Sample writeups:

• Triad City Beat
• DIYAI
• TextToLab

---

Independent reviews and voice AI comparison

Reviewers tend to agree on a few points:

• Voice quality sounds natural in blind tests, especially S2/S2-Pro
• Emotion and style control is above average for the price
• Generation and streaming are fast enough for production use
• Price-performance is strong vs ElevenLabs on API metering
• Feature set goes beyond basic TTS (ASR, separation, effects)

How it compares

Often mentioned next to ElevenLabs and OpenAI voice products. Fish Audio's edge in comparisons: realistic cloning, low latency, open-weight option, and API cost. ElevenLabs still wins some English-only quality tests; Fish Audio often wins on cost and control tags. Resemble is worth a look too if deepfake detection and voice security matter alongside generation.

Common themes in reviews: realism, speed, and dollars per million characters.

---

Company background and ecosystem

Fish Audio's commercial product sits on top of the Fish Speech open-weight work. Operated by Hanabi AI Inc. (founded 2024). The company announced a $52M seed and cites 8M+ builders on the homepage.

Open releases and a Discord community mean more tutorials, third-party integrations, and a real self-host path. You're not locked into only the hosted API if your team can run models themselves.

References:

• FoundersDatabase
• VoiceAISpace

---

Ethical and legal considerations

Voice cloning is useful and easy to misuse. Treat it like any powerful impersonation tool.

Consent

Get clear permission before cloning a real person's voice. Written consent is the minimum for anything commercial or public.

Deepfake risk

High-quality clones can fuel fraud and impersonation. Don't build deceptive apps. Platform terms matter; so do local laws.

Licensing

Free plan: personal, non-commercial per FAQ. Paid plans: commercial use on verified/cloned voices you own. Read terms before you monetize YouTube or client work.

Reasonable uses

Accessibility, localization, creative production, reducing studio cost for teams with rights to the source voice.

Bad uses

Scams, impersonation, non-consensual clones.

Simple rule: if it's a real person's voice, get consent in writing first.

Relevant pages:

• Voice clone
• Capabilities

---

Related tools on 101aitools

If Fish Audio isn't quite the fit, these voice AI tools are also listed on 101aitools:

• Fish Audio — the tool covered in this review; full listing with pricing and category tags
• Noiz AI — emoji-driven emotion control, 3-second voice cloning, and built-in video dubbing
• ElevenLabs — premium voice quality and cloning, often the benchmark in blind tests
• Murf AI — studio-style TTS built for business and e-learning voiceovers
• Play.ht — large voice and language library with commercial licensing on paid tiers
• Resemble — voice cloning paired with deepfake detection for security-conscious teams
• Speechify — consumer-focused TTS and reading assistant for accessibility use cases

---

FAQ

Is Fish Audio free to use? Yes, up to a point. The free plan gives ~8,000 credits a month (about 7 minutes of generation) with a 500-character cap per generation, and it's personal use only per the FAQ. Commercial use requires a paid plan.

How does Fish Audio pricing compare to ElevenLabs? On API usage, Fish Audio charges $15 per million UTF-8 bytes versus ElevenLabs' roughly $120–165 per million characters, based on third-party comparisons. On subscriptions, Fish Audio Plus starts at $15/mo ($5.50/mo billed annually). See ElevenLabs on 101aitools for a side-by-side on quality tradeoffs.

Can I self-host Fish Audio's models? Yes, for teams with GPU infrastructure. The S1/S2 model family is open-weight (Fish Speech, Apache 2.0), so you can run it privately instead of calling the hosted API. Most teams still use the API because it's simpler to operate and bill.

How fast is Fish Audio's voice cloning? Some flows work from as little as ~10–15 seconds of clean sample audio. Professional Voice Clone (PVC) slots, included on paid plans, take longer because they require live verification from the voice owner.

Is it safe to clone someone else's voice? Only with their written consent. Fish Audio's terms restrict cloning to voices you have rights to use, and misuse for impersonation or fraud violates both platform policy and, in most jurisdictions, the law.

What's the difference between S1, S2, and S2-Pro? S1 was the original commercial model. S2 improved realism, multilingual quality, and cloning. S2-Pro (and the current S2.1 Pro) adds better stability for long-form output and stronger emotional expressiveness. Word-level control (tagging a single word's prosody or pause) is an S2-family feature.