Audio Analysis
ASR (Speech-to-Text)
Transcribes speech to text with OpenAI's Whisper (or another Hugging Face speech model): one transcript per clip, or one row per segment or word with its start and end time.
When to use it
Call-centre analytics, meeting notes, voicemail to email, subtitles, or making recordings searchable — then hand the text to the Unstructured Text nodes.
In a pipeline
Audio File InputASR (Speech-to-Text)Diarization (Who Spoke When)Metadata Joiner
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Results | select | — | file |
| Language | text | — | Detect (or en, fr, hi…) |
| Task | select | — | transcribe |
| Model (Hugging Face id) | text | — | openai/whisper-small |
| Run on | select | — | auto |
| Time limit (seconds) | number | — | 900 |
Example
A voicemail, segment by segment
- Results=One row per segment, with times
- Model=openai/whisper-small
| File Name | Start (s) | End (s) | Transcript |
|---|---|---|---|
| vm-5521.wav | 0.6 | 4.2 | Hi, it's Ana from Northwind. |
| vm-5521.wav | 4.2 | 9.8 | Please call me back about the order. |
Worth knowing
- Leave Language empty to detect it; Translate turns any language into English text.
- whisper-small is a good balance on CPU; whisper-large-v3 is the most accurate, and wants a GPU.
- After a Diarization node (turns), each turn is transcribed on its own, giving who-said-what.
- Models run on the server in FlowDoo's Python and are downloaded once, on first use, into the data folder. A GPU is used when the server has one.