vapid olive
I am writing this after days of relentless efforts and numerous unsuccessful attempts to resolve a critical transcription issue with Eleven Labs.
Problem description:
The issue is that whenever I upload an audio file, the last part—approximately 10% of the transcription—is consistently missing. For example, a 3-minute audio yields a transcription only up to around 2 minutes and 30 seconds. This issue is persistent and occurs with every file, leaving the last segment of the audio untranscribed. AND ITS HAPPENING ONLY WITH TELUGU!
Here is a summary of my troubleshooting efforts:
Initially, I noticed the issue clearly at the end of the transcription. A strange character indicated a cut-off.
-
I then tried dividing the audio into smaller chunks and feeding them individually to Eleven Labs, but this created alignment issues between the audio timestamps and transcription.
-
If I feed the audio as a whole, the timestamps align perfectly, but the last portion remains missing.
-
Extending the audio length with silence padding did not resolve the problem.
-
Duplicating the audio (e.g., turning a 3-minute audio into 6 minutes by repeating it) bizarrely resulted in only the first 90% of the original segment being transcribed, after which it skips directly to the repeated segment and again only transcribes the first 90%.
-
Upon further research, I experimented with the "Scribe V1 Experimental" model, which, although improved slightly, led to another issue—transcriptions consistently defaulted to Devanagari (Hindi) script, even when inputting audio in other languages such as Telugu.
-
I also tried explicitly specifying file formats and details during upload and passed PCM data, but unfortunately, this yielded no improvements.
I have exhausted all conceivable methods and reached a point of deep frustration and desperation.
Please, please help me resolve this issue. Your prompt support and guidance would mean a great deal.
Thank you.