End-to-End Speech Benchmark

community

Activity Feed Request to join this org

AI & ML interests

Multi-domain, generalisable speech recognition.

Recent Activity

sanchit-gandhi authored a paper about 1 month ago

Voxtral TTS

sanchit-gandhi authored a paper 3 months ago

Voxtral Realtime

sanchit-gandhi authored a paper 10 months ago

Magistral

View all activity

sanchit-gandhi

authored a paper about 1 month ago

Voxtral TTS

Paper • 2603.25551 • Published Mar 26 • 61

sanchit-gandhi

authored a paper 3 months ago

Voxtral Realtime

Paper • 2602.11298 • Published Feb 11 • 26

sanchit-gandhi

authored 2 papers 10 months ago

Magistral

Paper • 2506.10910 • Published Jun 12, 2025 • 68

Voxtral

Paper • 2507.13264 • Published Jul 17, 2025 • 34

sanchit-gandhi

posted an update about 2 years ago

Post

Why does returning timestamps help Whisper reduce hallucinations? 🧐

Empirically, most practitioners have found that setting return_timestamps=True helps reduce hallucinations, particularly when doing long-form evaluation with Transformers’ “chunked” algorithm.

But why does this work?..

My interpretation is that forcing the model to predict timestamps is contradictory to hallucinations. Suppose you have the transcription:

The cat sat on the on the on the mat.

Where we have a repeated hallucination for “on the”. If we ask the model to predict timestamps, then the “on the” has to contribute to the overall segment-level timing, e.g.:

<|0.00|> The cat sat on the on the on the mat.<|5.02|>

However, it’s impossible to fit 3 copies of “on the” within the time allocation given to the segment, so the probability for this hallucinatory sequence becomes lower, and the model actually predicts the correct transcription with highest probability:

<|0.00|> The cat sat on the mat.<|5.02|>

In this sense, the end timestamp is of the opposite of the initial timestamp constraint they describe in Section 4.5 of the paper Robust Speech Recognition via Large-Scale Weak Supervision (2212.04356) → it helps the model remove extra words at the end of the sequence (rather than the initial timestamp which helps when the model ignores words at the start), but the overall principle is the same (using timestamps to improve the probability of more realistic sequences).

Leaving it open to you: why do you think timestamps reduces Whisper hallucinations?