OpenNMT / OpenNMT/CTranslate2

Whisper - Correct way to get prediction probability of each token and timestamp alignment

Open
#1,577 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
4.7k
Forks
536
Avg merge
12h 12m
Merged PRs (30d)
4

Description

I have spent time looking at the documentation but did not manage to find proper way to get the prediction probabilities of all tokens.
Also, how can i get the time-token alignment output from the model, with just the audio input features as input.
I did see the align function, but the function requires input features and the input text tokens - which does not seem to meet my need.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing the documentation and the existing align function, which currently takes input features and input text tokens. Determine the intended API for obtaining per-token prediction probabilities and time-token alignment from audio input features alone. Done means the requested outputs and supported inputs are clearly defined and exposed or documented.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.