# Whisper

> OpenAI Whisper speech-to-text - transcribe or translate audio.

OpenAI Whisper speech-to-text, built in on your FlyMyAI account. Give it a
public HTTPS audio URL and it returns recognized text you can feed into the
next step - an LLM reply, TTS, or an avatar pipeline. This is the preferred
STT tool for voice-note and avatar workflows.

## What it can do

| Method | What it does |
| --- | --- |
| `transcribe_audio` | Download a public HTTPS audio file (mp3, wav, m4a, webm, etc.) and return recognized speech as plain text. Set `task=transcribe` to keep the original language or `task=translate` to get an English translation. Optional `language` (ISO-639-1 code, omit for auto-detection) and `prompt` (style/vocabulary hint for names and product terms). Returns the text plus the detected language and audio duration when available. |

Pass `audio_url` as a public HTTPS link, such as an `agent_file.public_url`
from an uploaded voice message or a prior tool. A typical chain is: user
voice -> `agent_file.public_url` -> `transcribe_audio` -> compose a text
reply -> (optional) `elevenlabs.text_to_speech` -> (optional)
`musetalk.lip_sync`.

## How to get credentials

None - Whisper is a built-in tool. It runs on FlyMyAI infrastructure using
your account, with no API key to configure. Just enable it.

## Fields to fill in FlyMyAI

None.

## Troubleshooting

- **Input URL not fetched** - `audio_url` must be a direct public HTTPS
  link. Pre-upload via `download_link` or another agent tool if the source
  requires auth.
- **Wrong language detected** - pass an explicit `language` code (e.g. `en`,
  `ru`) instead of relying on auto-detection.
- **English output when you wanted the original language** - use
  `task=transcribe`; `task=translate` always returns English.