Skip to main content
The transcriber listens to the customer’s voice and converts it into text, so the agent’s brain (the LLM) can read it and reply. There are 6 providers to choose from. A few terms repeat across providers: Transcriber Model is the model used to convert the customer’s voice; Transcription Language is the language the customer’s speech is written down in; and Keywords are words the transcriber might mishear (brand names, short forms, special terms) that you add so it never gets them wrong. Soniox has a few extra features worth knowing:
  • Transcriber Language: in Soniox only, you can add multiple transcriber languages.
  • General Context: gives the agent a quick background of the call using simple key:value pairs, so it understands the conversation better before it even starts.
  • Text Context: tell Soniox what the call is about in one or two simple lines. The clearer you are, the fewer mistakes it makes.
  • Translation Terms: if Soniox keeps translating a word the wrong way, fix it here — you decide exactly how a word should appear, every single time.
  • Keywords: words Soniox might mishear — add them here so it never gets them wrong.

​
Check transcription quality

Review Audit Results

Use Transcription Quality, Transcription Issues, and Re-transcription to identify words the transcriber missed or captured incorrectly.