Speech to Text
Transcribe audio to text from uploads, recordings, or YouTube videos with word-level timestamps and speaker labels.
The STT (Speech to Text) tab transcribes audio into text. You can upload files, record from your microphone, paste a YouTube URL, or reuse audio you generated in the Lab.
Audio Source
The Audio Source panel provides these ways to load audio for transcription:
- Upload - Select an audio file from your computer
- Record - Record directly from your microphone, then click Stop
- YouTube - Paste a YouTube URL to extract and transcribe the audio (desktop app only)
- History - Click Use on any earlier TTS or Design generation in the History section below to load it for transcription
Language
Select the language of the audio from the Language dropdown, or leave it on Auto-detect to let the model identify the language automatically.
Speakers
Set Speakers to the number of people in the recording, from 1 to 6, or leave it on Auto. Each speaker gets a colored label in the results.
Speaker Detection
The Speaker detection menu picks the model that works out who is speaking:
- pyannote - Works on any computer. Most accurate when you set the number of speakers.
- Nemotron - Better at quick back-and-forth and short replies. Up to 8 speakers.
On the desktop app, Nemotron needs an NVIDIA graphics card. The menu only shows when more than one option is available.
ASR Model
Click the model name in the Results panel to switch between ASR (Automatic Speech Recognition) models:
| Model | Best for |
|---|---|
| Qwen3-ASR | Highest-accuracy transcription. A Quality switch picks Lower Quality (faster) or Higher Quality |
| Parakeet v3 | Fast transcription across many languages |
| Orukeet | Parakeet retrained for more accents and languages |
On the desktop app, Parakeet v3 and Orukeet need an NVIDIA graphics card.
Results
After clicking Transcribe, the transcription appears in the Results panel. With more than one speaker, each turn starts with that speaker's label.
Click any word to jump to that point in the audio.
If a turn has the wrong speaker, click its speaker label to move that turn, or every turn by that speaker, to someone else. To fix a single word, right-click it and move just that word or everything from there to the end of the turn. Pick an existing speaker or New speaker.
Click Copy text to copy the transcript, or export it as:
- SRT - Standard subtitle format, ready to use in video editors. Speaker labels carry over.
- JSON - Includes word-level timestamps and speakers for precise alignment and programmatic use
Use Cases
Generating captions
Transcribe video or podcast audio to create subtitles and captions.
Repurposing content
Convert spoken content into written form for blog posts, show notes, or social media.
Transcribing interviews and meetings
Set the number of speakers to get a transcript that shows who said what.
Transcription for voice cloning
Transcribe a reference audio clip to get an accurate transcript. The TTS tab also transcribes reference clips for you when you clone a voice.