Guide · Dify

Saidwire speech-to-text in Dify

Dify's OpenAI-API-compatible model provider can add a Speech2text model with your own API Base URL. Add Saidwire there, make it the workspace's default Speech-to-Text Model, and the microphone button in your Dify chat apps sends audio to Saidwire. You don't need to write any code.

you need
Dify	Dify Cloud, or a self-hosted Dify with plugins (1.0 or later). Checked against version 0.0.73 of the OpenAI-API-compatible plugin.
Saidwire API key	`sk_live_...` from saidwire.com/dev (120 free hours, no card)

1. Install the provider

In Dify, open Plugins and search the Marketplace for OpenAI-API-compatible (published by langgenius).
Click Install. If it's already installed, go to the next step. You can also get to it from Settings → Model Provider → Install model providers.

2. Add Saidwire as a Speech2text model

Open Settings → Model Provider. On the OpenAI-API-compatible card, click Add Model.

Fill in the form:

field	value
Model Type	`Speech2text`
Model Name	`saidwire-en`
Model display name	`Saidwire` (optional)
API Key	your Saidwire key, `sk_live_...`
Language	`en`, or leave it empty
Initial Prompt	leave it empty
API Base URL	`https://api.saidwire.com/v1`
model name for API endpoint	leave it empty

Click Save. To check the key, Dify transcribes its own 11-second sample clip. That uses 11 seconds of your free hours.

3. Make it the default Speech-to-Text Model

In Settings → Model Provider, open Default Models.
Under Speech-to-Text Model, pick saidwire-en (or the display name you gave it), then save.

4. Turn on voice input in an app

Open a chat app or chatflow and click Features.
Turn on Speech to Text ("Voice input can be used in chat"), then publish the app.
Click the microphone in the chat box and speak. The words go into the message box.

What Dify sends

For each recording, Dify sends one POST https://api.saidwire.com/v1/audio/transcriptions to Saidwire. The request is multipart and contains:

file: the recording. Chrome and Edge record WebM, and Safari records MP4.
model: the Model Name you entered.
language and prompt: sent only if you filled in those fields.

Dify reads back only the text field. We sent this exact request against the live API before publishing this page.

Troubleshooting

you see	fix
Credentials validation failed, with `401 Incorrect API key provided`	The key is wrong, has been revoked, or has a stray space.
`404` on save	Set API Base URL to exactly `https://api.saidwire.com/v1`. Dify adds `/audio/transcriptions` itself.
`400 ... unsupported_language`	Saidwire transcribes English only. Set Language to `en` or leave it empty.
Saidwire isn't listed under Speech-to-Text Model	The model was added with a different Model Type. Add it again with `Speech2text`.
`429`	Too many recordings at once for your key (trial keys allow 4), or your free hours are used up. The message says which.
Timeouts on self-hosted Dify	The plugin daemon container needs outbound HTTPS access to `api.saidwire.com`.

Good to know

This setup gives you text only. If you need word timestamps or SRT/VTT subtitles, call the endpoint directly. The API docs list every field.
Files can be up to 100 MB and 60 minutes. WAV, MP3, FLAC, Ogg/Opus, WebM and M4A/MP4 work. For M4A/MP4 files, the index has to be at the start of the file. Browser recordings already are. For ffmpeg output, add -movflags +faststart.
Saidwire processes the audio in memory and does not store it.
After the free hours, Saidwire costs $0.30 per audio hour, 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.