> ## Documentation Index
> Fetch the complete documentation index at: https://composio-preview.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Fish Audio

> Fish Audio provides speech synthesis, transcription, voice cloning, and conversational voice-agent APIs.

![Fish Audio logo](https://mintcdn.com/composio-preview/GL28i3tBlbEbBd-f/images/toolkits/fish_audio.svg?fit=max\&auto=format\&n=GL28i3tBlbEbBd-f\&q=85\&s=d01226df0f34e28cd37a6bf3963c2974)

* **Category:** audio & speech
* **Auth:** API\_KEY
* **Composio-managed OAuth available?** N/A
* **Tools:** 12
* **Triggers:** 0
* **Slug:** `FISH_AUDIO`
* **Version:** 20261101\_00
* **Instant:** Supported on the latest version. Support varies by tool.

## Tools

> [accordion] Configure Agent
>
> **Slug:** `FISH_AUDIO_CONFIGURE_AGENT`
>
> Partially update an agent's draft prompt, voice, and core conversation behavior without publishing it.
>
> #### Input Parameters
>
> **agent\_id** `string` *required*
>
> Agent ID to configure.
>
> **timezone** `string`
>
> Conversation timezone, up to 64 characters.
>
> **eagerness** `string ("relaxed" | "balanced" | "eager")`
>
> Response timing: relaxed, balanced, or eager.
>
> **record\_audio** `boolean`
>
> Whether future sessions should record audio.
>
> **first\_message** `string`
>
> Fixed greeting text when first\_message\_mode is fixed.
>
> **interruptible** `boolean`
>
> Whether callers may interrupt the agent while it speaks.
>
> **system\_prompt** `string`
>
> Replacement draft system prompt; Fish Audio limits it to 4000 tokens.
>
> **voice\_model\_id** `string`
>
> Replacement voice model ID returned by List Voice Models.
>
> **speaking\_language** `string ("en" | "ja" | "zh" | "ko" | "es" | "fr" | "de")`
>
> Speaking language: en, ja, zh, ko, es, fr, or de.
>
> **first\_message\_mode** `string ("off" | "fixed" | "prompt")`
>
> Greeting mode: off, fixed, or prompt.
>
> **first\_message\_prompt** `string`
>
> Instructions for generating the greeting when first\_message\_mode is prompt.
>
> **max\_duration\_seconds** `integer`
>
> Maximum session duration from 60 to 3600 seconds.
>
> **interruption\_sensitivity** `string ("low" | "balanced" | "high")`
>
> Interruption sensitivity: low, balanced, or high.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Create Agent
>
> **Slug:** `FISH_AUDIO_CREATE_AGENT`
>
> Create a private Fish Audio agent, optionally with an initial prompt, voice, and core conversation settings. The new agent remains a draft until separately published.
>
> #### Input Parameters
>
> **name** `string` *required*
>
> Agent name (1-120 characters).
>
> **description** `string`
>
> Agent purpose or operator-facing description (up to 2000 characters).
>
> **record\_audio** `boolean`
>
> Whether sessions should record audio.
>
> **first\_message** `string`
>
> Fixed initial greeting when first\_message\_mode is fixed.
>
> **system\_prompt** `string`
>
> Initial system prompt; Fish Audio limits it to 4000 tokens.
>
> **voice\_model\_id** `string`
>
> Voice model ID returned by List Voice Models.
>
> **speaking\_language** `string ("en" | "ja" | "zh" | "ko" | "es" | "fr" | "de")`
>
> Speaking language: en, ja, zh, ko, es, fr, or de.
>
> **first\_message\_mode** `string ("off" | "fixed" | "prompt")`
>
> Initial greeting mode: off, fixed, or prompt.
>
> **max\_duration\_seconds** `integer`
>
> Maximum session duration from 60 to 3600 seconds.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Delete Agent
>
> **Slug:** `FISH_AUDIO_DELETE_AGENT`
>
> Permanently delete a Fish Audio agent by ID; use only when deletion is explicitly intended.
>
> #### Input Parameters
>
> **agent\_id** `string` *required*
>
> ID of the agent to permanently delete.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Design Voice
>
> **Slug:** `FISH_AUDIO_DESIGN_VOICE`
>
> Generate original voice previews synchronously from a voice description, without cloning or saving a voice model.
>
> **Instant:** Supported on the latest version.
>
> **Instant pricing:** \$0.01 per successful voice-design request (1–4 candidates)
>
> #### Input Parameters
>
> **n** `integer`
>
> Number of voice candidates (1–4); the request is billed once.
>
> **seed** `integer`
>
> Optional deterministic generation seed.
>
> **speed** `number`
>
> Preview speaking speed multiplier, greater than zero and at most 3.
>
> **language** `string`
>
> Optional BCP-47 language hint, such as en or ja.
>
> **num\_step** `integer`
>
> Number of diffusion steps (1–128).
>
> **instruction** `string` *required*
>
> Describe the desired original voice; do not impersonate an identifiable person.
>
> **guidance\_scale** `number`
>
> Nonnegative classifier-free guidance scale.
>
> **reference\_text** `string`
>
> Optional preview text, up to 150 characters.
>
> **instruct\_guidance\_scale** `number`
>
> Nonnegative instruction guidance scale.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Get Account Balance
>
> **Slug:** `FISH_AUDIO_GET_ACCOUNT_BALANCE`
>
> Return the connected Fish Audio account's package allowance and API credit to distinguish package quota from paid credit. An optional team ID selects the team for the API-credit lookup; the package lookup remains for the connected account.
>
> #### Input Parameters
>
> **team\_id** `string`
>
> Optional team ID whose API credit should be checked; omit for the connected account's active workspace.
>
> **check\_free\_credit** `boolean`
>
> Ask Fish Audio to include current free-credit eligibility in the API-credit result.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Get Agent
>
> **Slug:** `FISH_AUDIO_GET_AGENT`
>
> Return an agent's workspace metadata together with its current draft conversation configuration.
>
> #### Input Parameters
>
> **agent\_id** `string` *required*
>
> Agent ID returned by List Agents or Create Agent.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Get Voice Model
>
> **Slug:** `FISH_AUDIO_GET_VOICE_MODEL`
>
> Get a Fish Audio voice model by ID and inspect whether it is trained and suitable for synthesis.
>
> #### Input Parameters
>
> **model\_id** `string` *required*
>
> Voice model ID returned by List Voice Models.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] List Agents
>
> **Slug:** `FISH_AUDIO_LIST_AGENTS`
>
> Search agents in the connected Fish Audio workspace, including draft or archived agents when requested.
>
> #### Input Parameters
>
> **search** `string`
>
> Case-insensitive search across agent names and descriptions.
>
> **page\_size** `integer`
>
> Maximum agents to return in this page (1-100).
>
> **next\_cursor** `string`
>
> Continuation cursor from a previous response. Omit for the first page; pass it back unchanged to fetch the next page.
>
> **include\_archived** `boolean`
>
> Whether to include archived agents.
>
> **publication\_state** `string ("all" | "live" | "draft")`
>
> Publication state to include: all, live, or draft.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] List Agent Sessions
>
> **Slug:** `FISH_AUDIO_LIST_AGENT_SESSIONS`
>
> List conversation sessions for the workspace or a specific agent, filtered by lifecycle status, caller, or creation time.
>
> #### Input Parameters
>
> **agent\_id** `string`
>
> Only return sessions belonging to this agent ID.
>
> **statuses** `array`
>
> Lifecycle statuses to include: pending, active, completed, failed, or unknown. A comma-separated string is also accepted.
>
> **page\_size** `integer`
>
> Maximum sessions to return in this page (1-100).
>
> **next\_cursor** `string`
>
> Continuation cursor from a previous response. Omit for the first page; pass it back unchanged to fetch the next page.
>
> **caller\_number** `string`
>
> Exact E.164 caller number for phone sessions.
>
> **created\_after** `string`
>
> Return sessions created after this ISO 8601 timestamp.
>
> **created\_before** `string`
>
> Return sessions created before this ISO 8601 timestamp.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] List Voice Models
>
> **Slug:** `FISH_AUDIO_LIST_VOICE_MODELS`
>
> Search public or workspace-owned Fish Audio voice models. Before synthesis, select an item whose type is tts; svc model IDs are not compatible with TTS.
>
> #### Input Parameters
>
> **tags** `array`
>
> Filter by one or more model tags. A single or comma-separated string is accepted.
>
> **title** `string`
>
> Filter models by title.
>
> **sort\_by** `string ("score" | "task_count" | "created_at")`
>
> Sort models by score, task\_count, or created\_at.
>
> **licensed** `boolean`
>
> When true, discover only voices whose rights Fish Audio has secured; ignored by the provider for owned\_only listings.
>
> **author\_id** `string`
>
> Filter public models by author ID; ignored when owned\_only is true.
>
> **languages** `array`
>
> Filter by one or more language codes such as en, es, or ja.
>
> **page\_size** `integer`
>
> Maximum models in this page (1-100).
>
> **owned\_only** `boolean`
>
> Return only models owned by the active workspace instead of public discovery models.
>
> **next\_cursor** `string`
>
> Continuation cursor returned by a previous call; omit for the first page.
>
> **title\_language** `array`
>
> Optional title-language filters.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Synthesize Speech
>
> **Slug:** `FISH_AUDIO_SYNTHESIZE_SPEECH`
>
> Generate downloadable speech synchronously using a Fish Audio model and voice IDs. With your own key the default is free S2.1; managed requests default to paid S2.1 Pro, require paid models and public, licensed, trained TTS voices, and are limited to 10,000 UTF-8 bytes and four voices.
>
> **Instant:** Supported on the latest version.
>
> **Instant pricing:** \$0.015 per 1,000 bytes of text; non-English characters use more bytes
>
> #### Input Parameters
>
> **text** `string` *required*
>
> Text to synthesize. For multiple voices, use \<|speaker:0|>, \<|speaker:1|>, and matching indexes in voice\_model\_ids.
>
> **model** `string ("s2.1-pro-free" | "s2.1-pro" | "s2-pro" | "s1")`
>
> Speech model. If omitted, defaults to free s2.1-pro-free when you connect your own Fish Audio key, otherwise to paid s2.1-pro. Without your own key, only paid models (s2.1-pro, s2-pro, s1) and public licensed voices are available. Preview models are excluded.
>
> **speed** `number`
>
> Speaking-rate multiplier from 0.5 to 2.0.
>
> **top\_p** `number`
>
> Nucleus-sampling diversity from 0 to 1.
>
> **format** `string ("mp3" | "wav" | "pcm" | "opus")`
>
> Output format: mp3, wav, pcm, or opus.
>
> **volume** `number`
>
> Output volume adjustment in decibels.
>
> **latency** `string ("normal" | "balanced" | "low")`
>
> Latency-quality tradeoff: normal, balanced, or low.
>
> **temperature** `number`
>
> Expressiveness from 0 to 1; higher values are more varied.
>
> **normalize\_text** `boolean`
>
> Normalize English and Chinese text, including numbers, before synthesis.
>
> **voice\_model\_ids** `array` *required*
>
> One voice model ID for normal speech, or multiple IDs in speaker-tag order for dialogue. A single string is accepted as a one-item list.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not

> [accordion] Transcribe Audio
>
> **Slug:** `FISH_AUDIO_TRANSCRIBE_AUDIO`
>
> Transcribe a previously uploaded audio file synchronously with Fish Audio. First upload through Composio files.upload in the SDK or the file-upload API; pass the returned file object, not a raw URL. Audio must be at most 64 MiB (a stricter platform limit may apply). Returns text, duration, optional segment timestamps and detected language. Long recordings may take several minutes.
>
> **Instant:** Supported on the latest version.
>
> **Instant pricing:** \$0.36 per audio hour, rounded up to the next second per successful request
>
> #### Input Parameters
>
> **audio** `object` *required*
>
> Upload the recording first using Composio files.upload in the SDK or POST /api/v3.1/files/upload/request followed by uploading the bytes. Pass the returned file object here. Maximum 64 MiB; a raw URL or local filename alone is not accepted. Use audio you own or have permission to transcribe.
>
> **model** `string ("transcribe-1" | "transcribe-1-pro")`
>
> Speech-to-text model; both models have the same published rate.
>
> **language** `string`
>
> Optional language hint; Fish Audio still auto-detects the language.
>
> **ignore\_timestamps** `boolean`
>
> Whether to omit precise timestamps; false can increase latency for short recordings.
>
> #### Output
>
> **data** `string` *required*
>
> Data from the action execution
>
> **error** `string`
>
> Error if any occurred during the execution of the action
>
> **successful** `boolean` *required*
>
> Whether or not the action execution was successful or not
