Speech Studio
- How to transcribe a recording, and pick how it is transcribed before anything is sent
- How to generate speech from text, and where the audio is kept
- How to find every transcription and generated voice in the History
The Speech Studio is where your team turns audio into text and text into audio. Open it from Speech Studio in the side menu — it is available to every member — and it opens as a tab next to your conversations. It has four tabs: Speech to text, Text to speech, History and Profiles.
Speech to text
Transcribing takes three steps, and nothing is sent until you press Transcribe:
- Audio — choose a file, pick one already in your storage, record with the microphone (press the button or the R key), or simply drag an audio file anywhere onto the studio, from any tab. The file is only prepared: you can Listen to it, Replace it or discard it.
- How to transcribe — pick a transcription profile. Profiles marked Separates speakers (such as AssemblyAI (platform), which every company has) label who said each part — use one for meetings, interviews and calls with more than one person. Each card also shows whether it runs on AutoTalk's platform account or your own key, and the longest file it accepts.
- Language — automatic detection works for most audio; fixing the language helps with short clips or strong accents.
The footer shows the chosen profile, the audio length and how it is billed. Press Transcribe to upload the audio and start the job. If the job is refused — for example because your credit ran out — the audio stays prepared, so pressing Transcribe again does not upload it twice.
Good to know:
- Speech to text needs permission to write to storage — the audio is stored for the transcription. Without it, the studio says so.
- Replacing a recording you prepared — with another file, a drop or a new recording — asks first: it hasn't been sent yet and is the only copy.
When the transcription is ready, Copy text and the .txt download sit at the top, with .srt, .vtt and .json in the menu next to it. The transcript below scrolls on its own, and you can search it or show only one speaker. ← New transcription brings you back to the composer.
Long files keep running in the background: you can leave the studio, and you'll get a notification when the job finishes.
Live mode transcribes while you speak, without a file; see the switch at the top of the composer.
Text to speech
Type what should be said, pick a voice profile and a format (MP3 plays everywhere; Opus is smaller), and press Generate speech.
Every generated clip is saved to your company storage automatically — the result shows Saved to storage, and Copy link shares it (that needs permission to share storage files). If it could not be saved (for example, the storage quota is full, or your role cannot write to storage), it shows Not saved; the audio still plays and can be downloaded. Your recent clips stay listed on the side, even after you close the studio.
History
The History tab lists the company's transcriptions and generated voices, newest first, wherever they came from — the studio (including sessions kept from its Live mode), Storage, the API, MCP-connected agents or workflows. Filter by Mine or Whole company, by type (Transcriptions / Generated voices), by origin, or search by file name or text, and press Load more to go further back. From a row you can open a transcription in the studio, download it, cancel or retry it, or play and download a generated voice.
A badge on the History tab counts your transcriptions that are still running.
Generated audio and transcripts follow your storage retention: once a file is removed — by retention, or deleted by hand in Storage — its row stays in the History marked File expired.
Profiles
The Profiles tab lists the company's transcription profiles (how audio becomes text: provider, model, whether speakers are separated) and voice profiles (how text becomes speech: provider, voice, model, speed). Each card shows whether it is the company default, whether it separates speakers, and whether it runs on AutoTalk's platform account or your own key.
The company owner can add a profile with New profile, change one with Edit (the same form is where it is removed), and pick the company default with Make default — except a profile that separates speakers, which channels can't use. Other members can see the profiles but not change them (the studio's link reads View profiles for them). Manage profiles in the speech-to-text and text-to-speech tabs brings you straight here. See Audio Transcription for what each transcription field does.
Billing
- Transcription on a platform profile bills the Speech-to-text meter; on your own key it is billed by your provider. See Audio Transcription.
- Speech generated on a platform voice bills the Text-to-speech meter. Saved clips count toward your storage like any other file.