French audio transcribed, translated and voiced in English

French audio transcribed, translated and voiced in English

An n8n workflow that carries speech through a full round trip. It turns a short French text into spoken French with ElevenLabs
and transcribes that audio back into text with OpenAI. An OpenAI chat model translates the text into English, and ElevenLabs reads
the English aloud. It works as a demo of an audio translation pipeline, and its parts can be swapped one at a
time.

What is inside

• workflow.json: the workflow as n8n exports it, twelve nodes. Seven of them run and five are sticky notes. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• The ElevenLabs voice ID belonged to the file's author. It is <YOUR_ELEVENLABS_VOICE_ID> now.
• The French paragraph in Set ElevenLabs voice ID and text was a passage from someone else's blog. We replaced it with a short
  text of our own that uses no quotation marks.
• The credential links are gone, and the file had no name, so we gave it one.
• Nothing else. The sticky notes, the requests and the prompt are as the file had them.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An ElevenLabs account and API key, and the ID of a voice in your voice library. Use a voice you may use.
• An OpenAI API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Pick or add a voice in your ElevenLabs voice lab and copy its ID. Open Set ElevenLabs voice ID and text and replace
   <YOUR_ELEVENLABS_VOICE_ID> with it. The brackets go.
3. In Generate French Audio, create a Header Auth credential. Name the header xi-api-key and give it your ElevenLabs key.
   Select the same credential in Translate English text to speech.
4. In Transcribe Audio, select an OpenAI credential. Select the same one in OpenAI Chat Model.
5. Edit the French text in Set ElevenLabs voice ID and text if you like.
6. Press Test workflow. Open the last node to listen to the English audio.

How a run goes

1. When clicking "Execute Workflow" starts the run by hand.
2. Set ElevenLabs voice ID and text sets two fields, voice_id and text.
3. Generate French Audio posts the text to ElevenLabs' text-to-speech address for that voice. It uses the model
   eleven_multilingual_v2, stability 0.5, similarity 0.5 and the header accept: audio/mpeg, and returns an audio file.
4. Transcribe Audio posts that file to OpenAI's /v1/audio/transcriptions as multipart/form-data, from the binary field
   data, with the model whisper-1. The reply holds the transcript in text.
5. Translate Text to English is a Basic LLM Chain, version 1.2, with the prompt Translate to English: and the transcript. Its
   model is OpenAI Chat Model, at temperature 0.
6. Translate English text to speech posts the English text to the same ElevenLabs voice, the way step 3 did, and returns the
   English audio as the last output. It escapes double quotes in the text first.

Change the look

• The text, the voice or the language: Set ElevenLabs voice ID and text, and the prompt in Translate Text to English.
• To translate a recording of your own: replace Set ElevenLabs voice ID and text and Generate French Audio with a node that
  gives your file as binary data named data, such as Read/Write Files from Disk.
• The target language: change English in the prompt of Translate Text to English, and drop the last node if you want text only.
• The speech model and voice settings: the request bodies in the two ElevenLabs nodes.
• The chat model: OpenAI Chat Model.

Good to know

• Generate French Audio builds its JSON by placing the text between quotation marks, with no escaping. A text that holds a
  double quote or a line break breaks that request. The second ElevenLabs node escapes double quotes, but a line break would break
  it too.
• Transcribe Audio sends whisper-1. OpenAI's deprecations page, read on 12 November 2026, shows it shutting down on 26 February
  2027, with gpt-live-transcribe or gpt-transcribe as the replacements. Check that the transcription address takes the new
  model before you switch.
• OpenAI Chat Model sets no model, so n8n uses its default for that node version, gpt-5-mini. The same page gives 11 December
  2026 for its shutdown, with gpt-5.6-terra as the replacement. Pick a model in the node before then.
• Both ElevenLabs requests send optimize_streaming_latency. ElevenLabs' documentation, read on 12 November 2026, marks it as
  deprecated but still lists it. Delete it from the query parameters if a request is refused.
• Speech made with a voice should be your own or one you have the right to use, and listeners should be told it is synthetic.
• Where the data goes: your text goes to ElevenLabs twice and to OpenAI as audio and as text, and the audio files stay in n8n's
  run data.
• Billing: ElevenLabs charges for the characters of both speech requests, and OpenAI charges for the audio minutes and the chat
  call.