Voice and Transcription

Dawn can turn speech into usable workspace text. Use voice input for short dictated prompts, and use transcription when an audio or video file should become a transcript attachment that can be searched, summarized, and used in follow-up work.

Voice Input vs Transcription

Dawn supports two speech workflows.

• Voice input: short dictation into the prompt composer. This is for speaking a message instead of typing it.
• Transcription: turning an uploaded or referenced audio/video file into a transcript file.

Use voice input when you want to ask Dawn something quickly. Use transcription when the recording itself is the work item.

Voice Input

Voice input is meant for short user prompts. Record the message, review the text that Dawn inserts into the composer, then send it like any other prompt.

Voice input is useful for:

• capturing a quick thought without typing;
• dictating a follow-up question;
• drafting a longer instruction while walking through an issue;
• using Dawn from a device where typing is inconvenient.

Voice input is not intended for long meetings or multi-speaker recordings. Upload the recording and transcribe it instead.

Transcribing Audio Or Video

To transcribe a file:

1. Attach an audio or video file to a Dawn thread, or mention an existing file.
2. Ask Dawn to transcribe it, or use the transcription prompt command when available.
3. Pick a language when you know the primary language, or leave language detection on automatic.
4. Wait for Dawn to create the transcript.
5. Open, download, summarize, or continue working with the generated transcript file.

Dawn can handle audio and video inputs by preparing the media for the transcription provider. The transcript is returned as a generated file so it can be reused later instead of being pasted only into one message.

Long Recordings

Long recordings may be split into smaller transcription segments so Dawn can stay within provider limits and recover from failures more safely.

Segmenting helps with:

• provider maximum-duration limits;
• large file handling;
• retrying a failed section without repeating the whole recording;
• producing a transcript for recordings that are too long for a single provider request.

When possible, Dawn uses speech-aware boundaries so segments cut near pauses instead of splitting speech mid-word. The final transcript should read as one document, not as a set of unrelated chunks.

Languages

If you know the primary language, choose it before transcribing. This helps the provider when a recording mixes languages or includes names, product terms, or accents that are easy to mishear.

Leave language on automatic when:

• you do not know the primary language;
• the recording is clearly single-language and provider auto-detection works well;
• you want the provider to infer the language without a hint.

For mixed-language conversations, choose the dominant language and include a prompt note about the other languages if they matter.

Speaker Labels

When supported by the selected provider, Dawn can preserve speaker-aware transcript structure. Speaker labels are best-effort. They are useful for meetings, interviews, and calls, but they may need review when the recording has overlapping speech, poor audio quality, or short segments with rapidly changing speakers.

Use the transcript as a working draft, then ask Dawn to clean up names or labels if the speakers are obvious from context.

Usage And Billing

Transcription can consume provider usage and workspace credits depending on the workspace configuration. Longer files generally cost more than short voice input.

Before relying on transcription at scale:

• confirm which provider the workspace uses;
• review the provider’s transcription limits;
• validate credit or usage tracking for your workspace;
• test with a representative recording before uploading many files.

Good Follow-Up Prompts

After a transcript is created, you can ask Dawn to work with it like any other file.

Summarize this transcript into decisions, risks, and action items.

Extract every customer request from the transcript into a table.

Turn this meeting transcript into release-planning notes.

Troubleshooting

• The upload is rejected: the file may be larger than the current upload limit. Split the file or use a smaller export.
• The provider rejects the file duration: split the recording into shorter parts or ask Dawn to transcribe through segmented processing.
• The transcript mixes languages badly: choose the dominant language and mention the other languages in the prompt.
• Speaker labels are wrong: ask Dawn to relabel speakers based on names or context in the transcript.
• No transcript file appears: refresh the thread and check the file surface. If it is still missing, retry the transcription.

Transcribe the attached recording and return a reusable transcript file.

Summarize this transcript into decisions, risks, and action items.