how convert to onnx
how convert to onnx
Hello, is there any update regarding ONNX deployment or any other deployment strategy for this model?
Hello, is there any update regarding ONNX deployment or any other deployment strategy for this model?
I've tried this model on production data and I personally dont recomment it at all. The idea is still good, but this model is bad in overall, way worse than Silero.
Hi @bizkst,
In which cases did you see this model perform worse than Silero?
I tested it for 3–4 hours in our deployment environment. The model mostly fixes filler-word false positives, but in some cases Silero detects speech correctly while this model misses it.
Could you share your experience? I’d be happy to hear your thoughts.
Could you share your experience? I’d be happy to hear your thoughts.
I've tested it in a full ASR pipeline with whisper\qwen models and it gets me about 2x worse WER. But I can agree that in some cases this model was better, but on my full dataset of long form audios it misses a lot real speech. I personally decided to not use it in prod env.
I have had the same experience, words are being missed by humaware that Silero woulnd't have missed.
I tested it on a french conference with heavy huming and I had false positives but mostly false negatives.
After taking a look at the dataset I realized that this model was not made for my use-cases.