Speech · Issues· 314 open

nemotron-labs-voicechat-11b: Training data

SubsamplingReductionModule pooling returns wrong lengths (factor 4 -> [6,5], factor 8 -> [0,0])

voice_agent: NemoSTTService's default model is misspelled ('nnvidia'), silently defaulting has_turn_taking to False

nvidia/nemotron-3.5-asr-streaming-0.6b: transcribe() fails with "ValueError: Unknown prompt key: 'None'"

Verify evals on Papers with Code

Mismatch of len(words) and len(word_confidence) in parakeet-tdt-0.6b-v3 transcription

ConvSubsampling forward broken: chunking=-1, striding_conv1d/dw_striding_conv1d, and vggnet all raise TypeError

Sortformer splits one speaker into two when their distance to the mic changes

Early-interruption augmentation can drop the relocated EOS token entirely

FastPitchModel_SSL / SSLDisentangler crash on validation: on_validation_epoch_end() missing 1 required positional argument: 'outputs'

No Streaming API for streaming sortformer?

missing punctuation marks about nvidia/nemotron-speech-streaming-en-0.6b

nemotron-labs-voicechat: Issues with the model's voice

Problem with computing drop_extra_pre_encoded when varying pre_encode_cache_size for SubSampling and VGG frontends

MagpieTTS raw-audio padding adds a spurious silent frame to exact-multiple-length audio