Skip to content

Audio transcription

Automatically convert conversation audio to text, with per-channel controls.

What it is

When audio is received or sent in a conversation, Interflow can generate a text transcript with AI (OpenAI Whisper). The text:

  • Appears with the audio message in the chat
  • Feeds the AI Agent and assistant context
  • Speeds up reading and searching history

Requirements

  • Channel with the desired transcription option enabled
  • Active OpenAI integration or organization AI credits balance

Without an OpenAI key or credits, audio is stored normally but without a transcript.

Per-channel settings

In Channels → edit the channel:

SettingDefaultScope
Transcribe customer audioOnAudio sent by the customer
Transcribe agent audioOffAudio sent by the agent in Interflow or the channel app
Transcribe group audioOffWhatsApp WAHA only, when groups are enabled

Existing channels

If the setting has not been saved yet:

  • Customer: stays on (previous behavior)
  • Agent and groups: stay off until you enable them

Audio sent from the agent's phone

For audio sent outside Interflow (from the phone WhatsApp) to be stored and transcribed:

  1. Enable upload of media sent via WhatsApp / external media on the channel
  2. Enable Transcribe agent audio

Without media upload, there is no file to transcribe.

Groups (WhatsApp WAHA)

By default, group audio is not transcribed (groups rarely use AI). To enable:

  1. Enable Enable groups on the WAHA channel
  2. Enable Transcribe group audio
  3. Keep customer and/or agent options as needed

Day-to-day use

  1. Configure the channel using the table above
  2. When an eligible audio is received or sent, wait for the transcript
  3. Read the Transcription block below the audio player
  4. Use the text in history and AI context for the next replies

Credits and cost

When the organization uses the Interflow key (AI credits), transcription consumes Audio Transcription credits based on duration. With the organization's own OpenAI key, usage follows the customer's OpenAI account.

Limitations

  • Transcription failure does not block sending or receiving the audio
  • Very long or low-quality audio may produce incomplete text
  • In WAHA groups, transcription only runs when Transcribe group audio is on

Documentation constantly being updated