Hugging Face's logo

STMicroelectronics
/
stft_tcnn

Model card Files
xet
Community
FBAGSTM commited on
Commit
8b5c9e5
·
verified ·
1 Parent(s): 0b6b66d

Release AI-ModelZoo-4.0.0

Browse files
Files changed (1) hide show
  1. README.md +15 -10
README.md CHANGED
@@ -1,3 +1,9 @@
 
 
 
 
 
 
1
  # STFT-TCNN
2
 
3
  ## **Use case** : `speech enhancement`
@@ -53,9 +59,9 @@ We also provide the original .yaml config file used to train the model. For deta
53
  Measures are done with default STEDGEAI configuration with enabled input / output allocated option.
54
 
55
  ### Reference **NPU** memory footprint
56
- |Model | Dataset | Format | Resolution | Series | Internal RAM | External RAM | Weights Flash | STM32Cube.AI version | STEdgeAI Core version |
57
- |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
58
- | [STFT-TCNN Medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/ST_pretrainedmodel_public_dataset/valentini/stft_tcnn_medium_sigmoid_257x40_qdq_int8.onnx) | valentini | Int8 | 257x40 | STM32N6 | 100.09 | 0.0 | 1599.39 | 10.2.0 | 2.2.0 |
59
 
60
  ### Reference **NPU** inference time
61
 
@@ -66,9 +72,9 @@ The figures listed in this table correspond to the version of ST Edge AI with th
66
  You can expect significant improvements once this issue is resolved.
67
 
68
 
69
- | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
70
- |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
71
- | [STFT-TCNN medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/ST_pretrainedmodel_public_dataset/valentini/stft_tcnn_medium_sigmoid_257x40_qdq_int8.onnx) | valentini | Int8 | 257x40 | STM32N6570-DK | NPU/MCU | 52.09 | 19.19 | 10.2.0 | 2.2.0 |
72
 
73
 
74
  ### Metrics on the Valentini dataset
@@ -83,8 +89,8 @@ We report five metrics :
83
 
84
  | Model | Format | Resolution | PESQ | STOI | SNR | SI-SNR | Waveform MSE |
85
  |-------|--------|------------|------|------|-----|--------|--------------|
86
- | [STFT-TCNN Medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/ST_pretrainedmodel_public_dataset/valentini/stft_tcnn_medium_sigmoid_257xsl_float.onnx) | float32 | 257x? | 2.480 | 0.931 | 18.190 | 18.104 | 1.136e-4 |
87
- | [STFT-TCNN Medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/ST_pretrainedmodel_public_dataset/valentini/stft_tcnn_medium_sigmoid_257xsl_qdq_int8.onnx) | int8 | 257x? | 2.372 | 0.932 | 18.190 | 18.100 | 1.109e-4 |
88
 
89
  ### Limitations
90
 
@@ -92,5 +98,4 @@ The models provided here typically have trouble denoising speech at SNRs beyond
92
 
93
  ## Retraining and Integration in a simple example:
94
 
95
- Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
96
-
 
1
+ ---
2
+ license: other
3
+ license_name: sla0044
4
+ license_link: >-
5
+ https://github.com/STMicroelectronics/stm32ai-modelzoo/blob/main/speech_enhancement/stft_tcnn/ST_pretrainedmodel_public_dataset/LICENSE.md
6
+ ---
7
  # STFT-TCNN
8
 
9
  ## **Use case** : `speech enhancement`
 
59
  Measures are done with default STEDGEAI configuration with enabled input / output allocated option.
60
 
61
  ### Reference **NPU** memory footprint
62
+ |Model | Dataset | Format | Resolution | Series | Internal RAM | External RAM | Weights Flash | STEdgeAI Core version |
63
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|-------------------------|
64
+ | [STFT-TCNN Medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/valentini/stft_tcnn_medium_sigmoid_257x40_qdq_int8.onnx) | valentini | Int8 | 257x40 | STM32N6 | 100.09 | 0.0 | 1578.39 | 3.0.0 |
65
 
66
  ### Reference **NPU** inference time
67
 
 
72
  You can expect significant improvements once this issue is resolved.
73
 
74
 
75
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version |
76
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|------------------------|
77
+ | [STFT-TCNN medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/valentini/stft_tcnn_medium_sigmoid_257x40_qdq_int8.onnx) | valentini | Int8 | 257x40 | STM32N6570-DK | NPU/MCU | 51.11 | 19.56 | 3.0.0 |
78
 
79
 
80
  ### Metrics on the Valentini dataset
 
89
 
90
  | Model | Format | Resolution | PESQ | STOI | SNR | SI-SNR | Waveform MSE |
91
  |-------|--------|------------|------|------|-----|--------|--------------|
92
+ | [STFT-TCNN Medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/valentini/stft_tcnn_medium_sigmoid_257xsl_float.onnx) | float32 | 257x? | 2.480 | 0.932 | 18.190 | 18.104 | 1.136e-4 |
93
+ | [STFT-TCNN Medium](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/speech_enhancement/stft_tcnn/valentini/stft_tcnn_medium_sigmoid_257xsl_qdq_int8.onnx) | int8 | 257x? | 2.372 | 0.932 | 18.190 | 18.100 | 1.109e-4 |
94
 
95
  ### Limitations
96
 
 
98
 
99
  ## Retraining and Integration in a simple example:
100
 
101
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)