card: same-session v3/Ultra/Phonon-2 LibriSpeech table
Browse files
README.md
CHANGED
|
@@ -67,10 +67,12 @@ top-64, fp16, iOS 17+), `Preprocessor.mlmodelc` and `parakeet_vocab.json` (ident
|
|
| 67 |
Full LibriSpeech, FluidAudio `asr-benchmark`, M5 Pro, both models run back to back with the default (ANE) encoder. WER is
|
| 68 |
corpus-level (total edit distance over total reference words); RTFx = total audio / total processing time.
|
| 69 |
|
| 70 |
-
| Set | v3 |
|
| 71 |
-
|---|---
|
| 72 |
-
| test-clean (2620 files) | **2.
|
| 73 |
-
| test-other (2939 files) |
|
|
|
|
|
|
|
| 74 |
|
| 75 |
v3 is the more accurate model on LibriSpeech by 0.20 / 0.50 points, reproducing the upstream card's own gap to its
|
| 76 |
teacher (+0.20 / +0.79 under the Open ASR Leaderboard protocol, where Phonon-2 in turn beats v3 on AMI meetings and
|
|
|
|
| 67 |
Full LibriSpeech, FluidAudio `asr-benchmark`, M5 Pro, both models run back to back with the default (ANE) encoder. WER is
|
| 68 |
corpus-level (total edit distance over total reference words); RTFx = total audio / total processing time.
|
| 69 |
|
| 70 |
+
| Set (ANE) | v3 | Ultra | Phonon-2 |
|
| 71 |
+
|---|---|---|---|
|
| 72 |
+
| test-clean (2620 files) WER | 2.27 % | **2.13 %** | 2.47 % |
|
| 73 |
+
| test-other (2939 files) WER | 4.12 % | **3.81 %** | 4.62 % |
|
| 74 |
+
| test-clean RTFx | 149–152× | 151× | **159×** |
|
| 75 |
+
| test-other RTFx | 138× | 142× | **146×** |
|
| 76 |
|
| 77 |
v3 is the more accurate model on LibriSpeech by 0.20 / 0.50 points, reproducing the upstream card's own gap to its
|
| 78 |
teacher (+0.20 / +0.79 under the Open ASR Leaderboard protocol, where Phonon-2 in turn beats v3 on AMI meetings and
|