Guide · n8n

Saidwire speech-to-text in n8n

n8n's built-in OpenAI node can transcribe audio, and its OpenAI credential has a Base URL field. Point that field at Saidwire and the node's Transcribe a recording action runs on Saidwire with no other changes.

you need
n8n	Cloud or self-hosted, with a recent version whose OpenAI credential shows a Base URL field
Saidwire API key	`sk_live_...` from saidwire.com/dev (120 free hours, no card)

Saidwire community node

In n8n, open Settings → Community Nodes and install n8n-nodes-saidwire. Self-hosted:

npm install n8n-nodes-saidwire

npm package · Source on GitHub · The steps below use n8n's built-in OpenAI node with our base URL and work too.

1. Create the credential

In n8n, open Overview, click Create at the top-right, and choose Credential. (You can also create it from inside the node in step 2, under Credential to connect with.)
Search for OpenAI and select it.
Fill in the fields:
field value

API Key your Saidwire key, sk_live_...

Organization ID (optional) leave blank

Base URL https://api.saidwire.com/v1 (no trailing slash)

Add Custom Header off
Click the credential's name at the top and rename it to something like Saidwire, so it is not confused with an OpenAI key.
Click Save. n8n tests the credential by listing models, and Saidwire answers with saidwire-en, so the test passes when the key is valid.

field	value
API Key	your Saidwire key, `sk_live_...`
Organization ID (optional)	leave blank
Base URL	`https://api.saidwire.com/v1` (no trailing slash)
Add Custom Header	off

2. Add the transcription step

Start with a node that outputs an audio file as binary data, for example Read/Write Files from Disk, an HTTP Request that downloads a file (Response Format: File), a Form trigger with a file field, or a Telegram trigger set to download voice messages.
Click + after it, search for OpenAI, and in its list of actions choose Transcribe a recording (under the audio actions).

Set the parameters:

parameter	value
Credential to connect with	the `Saidwire` credential from step 1
Resource	`Audio`
Operation	`Transcribe a Recording`
Input Data Field Name	the binary field holding the audio, usually `data`
Options → Language of the Audio File	`en` (optional; other languages are rejected)
Options → Output Randomness (Temperature)	leave unset (it has no effect)

Click Execute step. The output is {"text": "...", "usage": {"type": "duration", "seconds": ...}}. Use {{ $json.text }} in the next node.

The node always sends model=whisper-1. Saidwire accepts any model name, so this is fine.

Timestamps or subtitles: use HTTP Request

The OpenAI node returns plain text only. For word timestamps, segments, or an SRT/VTT file, call the endpoint with an HTTP Request node and reuse the same credential:

parameter	value
Method	`POST`
URL	`https://api.saidwire.com/v1/audio/transcriptions`
Authentication	`Predefined Credential Type`
Credential Type	`OpenAI`, then pick the `Saidwire` credential
Send Body	on
Body Content Type	`Form-Data`
Body field 1	Type `n8n Binary File`, Name `file`, Input Data Field Name `data`
Body field 2	Type `Form Data`, Name `model`, Value `saidwire-en`
Body field 3	Type `Form Data`, Name `response_format`, Value `verbose_json` (or `srt`, `vtt`, `text`)
Body field 4 (optional)	Type `Form Data`, Name `timestamp_granularities[]`, Value `word`

For srt, vtt and text, set Options → Response → Response Format to Text. The full list of fields and formats is in the API docs.

Troubleshooting

you see	fix
Credential test fails, or `404`	Base URL must be exactly `https://api.saidwire.com/v1`, with no trailing slash.
`401 Incorrect API key provided`	The key is wrong, revoked or has a stray space.
`400 ... language`	Saidwire transcribes English only. Remove the language option or set `en`.
`400 ... file`	The binary field is empty or not audio. Check Input Data Field Name against the previous node's output.
`429`	Too many files at once for your key (trial keys allow 4), or your free hours are used up. Lower the batch size, or check the message.

Good to know

Files up to 100 MB and 60 minutes. WAV, MP3, FLAC, Ogg/Opus, WebM and M4A/MP4 work.
Saidwire processes the audio in memory and does not store it.
$0.30 per audio hour after the free hours, billed per second.

Stuck? Email support@saidwire.com with the x-request-id from the failed call.

OpenAI is a trademark of OpenAI. Open WebUI, n8n, LiteLLM, Dify, Flowise, Vercel and the AI SDK belong to their respective owners. Saidwire is not affiliated with OpenAI or with these projects.