Pipe Can Now Transcribe Your Recordings
The Pipe Recording Platform can now automatically transcribe your audio and video recordings.
We've integrated with Replicate and currently use a variant of OpenAI's Whisper Large V3 model, more specifically, incredibly-fast-whisper, to generate transcriptions across a large pool of languages quickly.
The transcription feature is opt-in.
A Quick Look Behind the Scenes
Audio transcription runs as part of Pipe's existing processing pipeline.
First, we extract the audio from the recording as a highly optimized audio file prepared for transcription. This process is independent of Pipe's existing audio extraction feature.
Once audio is ready, we send it to Replicate for transcription. The transcription file will be a JSON file - more details below -that we push to our storage or your storage together with all the other recording files.
For most recordings, transcription takes only a few seconds. Very long recordings can take up to a few minutes. According to the model's own benchmark, it transcribes 150 minutes of audio in 100 seconds.
It supports 100 languages, and automatically detects the language.
What Happens to Your Audio?
Replicate retains the uploaded audio only while transcription is running and deletes it once transcription is complete. The resulting transcription data remains on Replicate for up to one hour after transcription finishes.
Replicate does not use customer-provided audio data to train its models.
Transcription Output
The resulting transcription file will be in JSON format and named STREAMNAME_transcript.json. It contains contains timestamped chunks, the full transcript text and metadata about the transcription job. The chunks/full text include punctuation. You can find the complete JSON schema in the transcription documentation.
Turning ON Audio Transcription
To enable transcription:
- Log in to your Pipe account dashboard.
- Navigate to the Transcoding Engine page.
- Scroll down to the Audio Transcription section.
- Enable Transcribe new recordings.
- Save
Once enabled, new recordings will be automatically transcribed as part of the processing pipeline.
Beta Feature
Audio transcription is currently a beta feature. Some aspects of the feature, including the transcription output(s), may change as we continue to improve it.
If you try it out, we’d love to hear your feedback. Feel free to contact us with any comments or suggestions.
Pricing
Audio transcription is currently available as a complimentary service under fair use. The feature is available today during the trial and with all paid subscriptions. This may change in the future.
