Fish Audio by AceDataCloud — n8n community node

Turn text into Fish Audio speech, use reference voices, discover voice models, and track synthesis tasks in n8n with AceDataCloud.

Package: @acedatacloud/n8n-nodes-fish · License: MIT · Node: Fish Audio by AceDataCloud

Maintained by Ace Data Cloud, connecting to AceDataCloud's API. This is not an official package from the model developer. A direct provider subscription does not supply an AceDataCloud API credential.

What you can build

• Turn articles, notifications and scripts into spoken audio.
• Find a voice in the public library or reuse a known voice ID.
• Use a short reference recording for one-time voice matching.

Installation

In self-hosted n8n, open Settings → Community nodes → Install, enter @acedatacloud/n8n-nodes-fish, review the package and install. To reproduce this release exactly, use @acedatacloud/n8n-nodes-fish@0.1.0.

See n8n's installation guide. n8n Cloud requires verified community nodes. npm publication and verification are separate; check the n8n node panel for current Cloud availability.

Operations

Credentials

1. Open AceDataCloud applications, sign in and create or select an application.
2. Copy the application's API token. One AceDataCloud application token can access the platform services available to your account.
3. In n8n, create a Fish Audio by AceDataCloud API credential, paste the API Token, save it and confirm the credential test passes.
4. Select that credential on each Fish Audio node in your workflow.

The token is masked and stored in n8n's credential store. The node sends Bearer authentication only to https://api.acedata.cloud. Its credential test queries an empty task batch and does not generate media. Do not use a platform management token.

Quick start and examples

Import Generate and wait, assign credentials to Create and Get Task, and edit the generation parameters. One manual execution submits once, then queries every 15 seconds. The loop has a 30-minute deadline and explicit failure handling. A deadline stops waiting; it does not cancel a submitted service task.

Import AI Agent task tool to use the node as a read-only tool. Replace the sample task ID, configure the chat model and select the service credential. The agent queries a task once; it does not create media.

Additional examples are in examples/. They contain no credentials and no pinned or fabricated output.

Example output

Open a real audio result generated through this node with s2-pro during live validation.

Parameters and limits

• TTS models: s1, s2-pro and s2.1-pro; model selection is sent in the API model header.
• Output: MP3, WAV or PCM. PCM results use a WAV container URL.
• Choose one voice source: default, one or more voice IDs, or a single public HTTPS reference recording with its exact transcript. Use reference audio you own or have permission to use.
• Options include speed, volume, normalization, sample rate, latency and MP3 bitrate.
• Voice Search returns at most 100 results per page. Use Page Number and filters for larger searches; it does not download the entire voice library.
• Persistent voice-model creation is not included in this release.

Output and task states

With Simplify enabled:

• Generation returns taskId, status: "submitted", finished: false, successful: null and traceId.
• Task queries return taskId, status (processing, succeeded or failed), finished, successful, data, error, traceId and cost.
• Final media URLs are inside data. Fish TTS responses with a top-level audio_url are exposed as data.audio_url.
• A submission acknowledgment is not a finished result. Continue only when finished is true and successful is true.

Disable Simplify to receive the API response unchanged. Batch queries emit one item per task, preserving the relationship to the originating n8n input item.

Billing, retries and errors

Generation uses your AceDataCloud balance. Check current API documentation and pricing before execution. cost.amount is in Credits, not USD; the conversion depends on your current package rate.

Each incoming n8n item sends one generation request. Creation is not retried automatically and the node does not substitute a different model. Re-running Create, enabling retry-on-fail, or restarting the workflow can create another paid task. After a timeout, retrieve the existing task where possible.

HTTP/authentication/parameter errors fail the node with the API's error context. Completed failed tasks are represented as finished: true, successful: false and an error, so downstream nodes can handle them. n8n's continue-on-fail option emits an error item and proceeds with later input items.

Data handling and compatibility

Only configured text, model choices, media references and query parameters are sent to AceDataCloud. Input URLs must be accessible to the API. The node has no filesystem access, environment-variable access or telemetry, and no runtime dependencies beyond n8n's workflow API.

Built with the official @n8n/node-cli, strict community-package rules and Node.js 24. CI loads the package in n8n 2.42.3. Older n8n versions are not part of this release's compatibility test matrix. Icons are bundled from the existing Studio/service catalog; see asset provenance.

Development

pnpm install --frozen-lockfile
pnpm run build
pnpm run lint
pnpm test

Programmatic nodes validate conditional media payloads, normalize asynchronous task responses and preserve input-item links. All service requests use n8n's authenticated HTTP helper. Unit tests mock the HTTP boundary; they do not incur generation charges.

Releases are built from merged source through GitHub Actions with npm publish --provenance. Publication runs build, lint and tests. Keep published versions immutable and update CHANGELOG.md for new releases.

Resources and support

• API documentation
• AceDataCloud Studio
• Report an issue
• n8n community-node documentation

Include the package version, n8n version, operation and a redacted error when reporting issues. Never include API tokens or private media.