File Transcription

Speakly transcribes audio and video files you already have: recorded meetings, interviews, voice memos, lecture videos.

How to use

1. Open Transcribe Media → New Transcription and drop an audio/video file or click Choose File.
2. Choose Transcription engine and Model for this import. Configure keys and download local models in Settings → Transcription. Google and ElevenLabs use file transcription models, not their live streaming variants.
3. Choose Speech language or Auto-detect. Automatic detection ignores the global speech language. Your media choices are remembered separately from dictation and Meetings.
4. Set Identify speakers, the number of speakers (Auto or 1–8, counting everyone), and Keep the audio, then click Start Transcription.

Supported formats and limits

Audio: MP3, WAV, M4A, OGG, WEBM, FLAC, AAC, WMA. Video: MP4, MOV, AVI, MKV. Files must contain a readable audio track and last at most 2 hours. Unsupported formats show an error. Files over 8 minutes are automatically divided into chunks; timestamps stay on the full file timeline.

Progress and failures

The dialog shows preparation, transcription, speaker identification and saving, with progress and part counts where available. Controls stay locked while processing. If speaker analysis fails, the transcript is kept with a warning. If local storage fails, retry saving or choose Save transcript only. When post-processing is configured, a short title is generated from the transcript, like in Meetings; the file name stays visible under it. Click the title to rename it.

Identify and edit speakers

Identification runs locally with the same one-time 28 MB voice model used by Meetings. It starts enabled when that model is installed; otherwise download it in the dialog. Click a speaker name to rename them or assign that line to another speaker. With the audio kept, the Who is who? row plays each speaker's longest line, so you can put a name to a voice before renaming. You can also merge speakers. Copied text and AI tools use the updated names. Splitting mixed passages requires real word timestamps; uncertain passages can remain unidentified instead of receiving guessed labels.

Keep and reprocess audio

Keep the audio starts enabled. It stores a compressed audio copy locally, including audio extracted from videos. Play the recording or click a timestamp to replay a line. With retained audio, Re-transcribe replaces text in the existing intervals using another engine/model/language; Identify speakers again reruns the voice analysis; if you set a number of speakers and fewer distinct voices are found, the transcript keeps the ones found and a notice says so. Names and manual speaker corrections are preserved. A failed reprocessing operation keeps the existing transcript. Both actions ask for confirmation first, and closing the transcript while one runs cancels it.

Local storage and providers

Only the local transcription engine processes speech entirely on your computer. BYOK engines send audio to the provider selected in the import dialog. Speaker analysis and retained copies stay local. With retention off, temporary audio is discarded after processing, while text and speaker labels remain. Deleting a transcript also deletes its retained audio. Your original file is never modified or deleted. You can also remove all retained audio through Settings → Data and Privacy → Saved audio files → Delete, keeping the transcripts.

Older transcripts remain readable. Imports without retained audio cannot be replayed or reprocessed; import the original again to gain those features. These options apply to imported files; video links keep their separate workflow.