timestamps
Hi, thanks for training this — impressive WER (4.13%) for a low-resource
language like Sorani Kurdish!
Question: does this model support word-level or sentence-level timestamps
(forced alignment), either built-in or via a companion model? I'm using it
for a video dubbing pipeline where I need accurate start/end times for
each sentence, not just the transcribed text.
If it doesn't currently, do you know of a forced-alignment approach that
works well with this model's output (e.g., CTC-segmentation, a
wav2vec2-based aligner, or similar)?
Hello, yes, it supports word-level timestamps
You can use the model from our platform: https://pawan.krd/models/pkrd/asr-ku-large
We also support separating different speakers in the audio
can you give me access to download this model ?
Sorry, this isn't an open or research model, you can only access it through our platform.