Lakoc commited on
Commit
bc48eb9
·
verified ·
1 Parent(s): df53a3a

Correct the description of the pseudo-labeled stage

Browse files
Files changed (1) hide show
  1. README.md +5 -1
README.md CHANGED
@@ -11,6 +11,10 @@ pipeline_tag: automatic-speech-recognition
11
  library_name: nemo
12
  base_model: nvidia/parakeet-tdt-0.6b-v3
13
  tags:
 
 
 
 
14
  - speech
15
  - automatic-speech-recognition
16
  - multi-talker-asr
@@ -83,7 +87,7 @@ Three stages, each starting from the previous one:
83
  | stage | data | hours |
84
  |---|---|---|
85
  | 1. synthetic | [FastMSS](https://arxiv.org/abs/2605.15442) mixtures of up to 8 LibriSpeech speakers (2k h) + Fisher conversations restored with Sidon and force-aligned (2k h) | 4.0k |
86
- | 2. pseudo-labeled | multi-speaker YODAS2 conversations, diarized with streaming Sortformer and transcribed by DiCoW v3.3, filtered for language ID, hallucinated repetitions and cross-speaker leakage | 4.6k |
87
  | 3. real | NOTSOFAR-1 and AMI single-distant-microphone training sets | 115 |
88
 
89
  Training conversations are capped at 8 speakers. The diarization mask is randomly corrupted during
 
11
  library_name: nemo
12
  base_model: nvidia/parakeet-tdt-0.6b-v3
13
  tags:
14
+
15
+
16
+
17
+
18
  - speech
19
  - automatic-speech-recognition
20
  - multi-talker-asr
 
87
  | stage | data | hours |
88
  |---|---|---|
89
  | 1. synthetic | [FastMSS](https://arxiv.org/abs/2605.15442) mixtures of up to 8 LibriSpeech speakers (2k h) + Fisher conversations restored with Sidon and force-aligned (2k h) | 4.0k |
90
+ | 2. pseudo-labeled | multi-speaker YODAS2 conversations in the 23 languages Parakeet-v3 supports, automatically diarized and transcribed by DiCoW v3.3 conditioned on that diarization | 4.6k |
91
  | 3. real | NOTSOFAR-1 and AMI single-distant-microphone training sets | 115 |
92
 
93
  Training conversations are capped at 8 speakers. The diarization mask is randomly corrupted during