mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-10 02:22:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

update multi-hans whisper-qwen-1.5b results (#1657)

Browse Source
This commit is contained in:
Yuekai Zhang 2024-06-19 11:10:31 +08:00 committed by GitHub
parent 2e05663fbb
commit ff2bef9e50
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 27 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
egs/speech_llm/ASR_LLM/RESULTS.md
View File

@ -2,12 +2,31 @@
### whisper_llm_zh finetuning results ### whisper_llm_zh finetuning results
| Training Dataset | Speech Encoder | LLM | Projector |Comment | CER | |Model| Training Dataset | Speech Encoder | LLM | Projector |
| -------------------------| ----------------|------|--------------------------------------------------|-----|--| |-| -------------------------| ----------------|------|---------------|
| Aishell1 | whisper-large-v2-aishell1-ft, freeze| Qwen2-1.5B-Instruct, LoRA | Linear, 8x downsample| [yuekai/icefall_asr_aishell_whisper_qwen2_1.5B](https://huggingface.co/yuekai/icefall_asr_aishell_whisper_qwen2_1.5B) | Aishell1 Test 3.62% | |[yuekai/icefall_asr_aishell_whisper_qwen2_1.5B](https://huggingface.co/yuekai/icefall_asr_aishell_whisper_qwen2_1.5B) | Aishell1 | whisper-large-v2-aishell1-ft, freeze| Qwen2-1.5B-Instruct, LoRA | Linear, 8x downsample|
<!-- | Multi-hans-zh | whisper-large-v2-multi-hans-ft, freeze| Qwen2-1.5B-Instruct, LoRA | Linear, 8x downsample| WIP || | [yuekai/icefall_asr_multi-hans_whisper_qwen2_1.5B](https://huggingface.co/yuekai/icefall_asr_multi-hans_whisper_qwen2_1.5B) |Multi-hans-zh | whisper-large-v2-multi-hans-ft, freeze| Qwen2-1.5B-Instruct, LoRA | Linear, 8x downsample||
| Multi-hans-zh | whisper-large-v2-multi-hans-ft, freeze| Qwen2-7B-Instruct, LoRA | Linear, 8x downsample| WIP || -->
CER Details:
| Model | [yuekai/icefall_asr_aishell_whisper_qwen2_1.5B](https://huggingface.co/yuekai/icefall_asr_aishell_whisper_qwen2_1.5B) | [yuekai/icefall_asr_multi-hans_whisper_qwen2_1.5B](https://huggingface.co/yuekai/icefall_asr_multi-hans_whisper_qwen2_1.5B) |
|-------|------------------------------------------------|----------------------------------------------------|
| Split | Greedy Search | Greedy Search |
| aishell-1 dev | - | 0.66 |
| aishell-1 test | 3.62 | 0.68 |
| aishell-2 dev | - | 2.67 |
| aishell-2 test | - | 2.94 |
| aishell-4 test | - | 16.20 |
| alimeeting eval | - | 30.86 |
| alimeeting test | - | 40.50 |
| magicdata dev | - | 2.50 |
| magicdata test | - | 1.70 |
| kespeech-asr dev phase1 | - | 6.22 |
| kespeech-asr dev phase2 | - | 2.18 |
| kespeech-asr test | - | 6.59 |
| WenetSpeech dev | - | 4.59 |
| WenetSpeech test_meeting | - | 6.41 |
| WenetSpeech tes_net | - | 6.63 |
| SPEECHIO Avg 001-026 | - | 4.80 |
Command for training is: Command for training is:
```bash ```bash
pip install -r whisper_llm_zh/requirements.txt pip install -r whisper_llm_zh/requirements.txt

7
egs/speechio/ASR/local/normalize_results.py
View File

@ -16,12 +16,11 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
""" """
This file uses whisper and zipformer decoding results to generate fusion decoding results. This file uses speech io offcial pipline to normalize the decoding results.
Since whisper model is more likely to make deletion errors and zipformer model is more likely to make substitution and insertion errors, https://github.com/SpeechColab/Leaderboard/blob/master/utils/textnorm_zh.py
we trust whisper model when it makes substitution and insertion errors and trust zipformer model when it makes deletion errors.
Usage: Usage:
python whisper_zipformer_fusion.py --model-log-dir ./whisper_decoding_log_dir --output-log-dir ./results_norm python normalize_results.py --model-log-dir ./whisper_decoding_log_dir --output-log-dir ./results_norm
""" """
import argparse import argparse