Skip to main content
Open APIs & Integrations → Speech to Text or Text to Speech to experiment with the same speech backends your agents use. Playgrounds help you pick providers and settings with evidence instead of guesses. Speech access is plan-gated: workspaces without voice access see these pages locked with an upgrade message.

​
Test speech-to-text

  1. Open the Speech to Text playground.
  2. Provide audio by uploading a file (WAV, MP3, FLAC, or OGG, up to 100 MB) or recording from your microphone.
  3. Run transcription and review the text segments, including speaker separation (diarization) where available.
  4. Download the transcript (.txt) to compare providers side by side.
The page shows ready-to-copy JavaScript, Python, and cURL samples for the same request, so a successful test converts directly into integration code. Each run reports credits deducted in the response header.

​
Test text-to-speech

  1. Open the Text to Speech playground.
  2. Enter the exact sentence your agent will speak - short, punctuation included.
  3. Choose a voice, then adjust pace and sample rate.
  4. Play the generated WAV and judge it on a phone speaker, not just laptop audio.
Keep the voice that stays clear at fast pace and under background noise; note its provider and model for the agent’s Voice & Audio panel.

​
Costs and history

  • Playground runs consume credits (stt-playground / tts-playground entries) - check Credit Usage after a test batch.
  • Every request is logged with its details under API Logs; open a row to replay what was sent and received.
  • If results differ between playground and live calls, compare provider, model, pace, and audio conditions before changing the agent - the playground is clean audio, live calls are not.
Decide the voice in the playground, then configure the agent once. Changing voices after launch forces retesting every conversation.

Configure the voice

Apply the winning voice to your agent.

Review API history

Inspect each playground request.