MLECO-1883: Updating wav2letter model
Using the new pruned wav2letter model from Arm Model Zoo.
The new model when optimised by Vela, produces a tflite
file ~10 MB smaller than the current.
Change-Id: I4ab6007c5b6111f41d8097e29b2af6cde2abc457
diff --git a/docs/documentation.md b/docs/documentation.md
index 9ec73a3..7f8fbf9 100644
--- a/docs/documentation.md
+++ b/docs/documentation.md
@@ -10,7 +10,7 @@
- [Deployment](#deployment)
- [Implementing custom ML application](#implementing-custom-ml-application)
- [Testing and benchmarking](#testing-and-benchmarking)
- - [Memory considerations](#memory-considerations)
+ - [Memory Considerations](#memory-considerations)
- [Troubleshooting](#troubleshooting)
- [Appendix](#appendix)
@@ -184,7 +184,7 @@
- [Mobilenet V2](https://github.com/ARM-software/ML-zoo/blob/master/models/image_classification/mobilenet_v2_1.0_224/tflite_uint8).
- [DS-CNN](https://github.com/ARM-software/ML-zoo/blob/master/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8).
-- [Wav2Letter](https://github.com/ARM-software/ML-zoo/blob/master/models/speech_recognition/wav2letter/tflite_int8).
+- [Wav2Letter](https://github.com/ARM-software/ML-zoo/tree/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8).
- [Anomaly Detection](https://github.com/ARM-software/ML-zoo/raw/7c32b097f7d94aae2cd0b98a8ed5a3ba81e66b18/models/anomaly_detection/micronet_medium/tflite_int8/ad_medium_int8.tflite).
When using Ethos-U55 NPU backend, the NN model is assumed to be optimized by Vela compiler.
diff --git a/docs/quick_start.md b/docs/quick_start.md
index abf8f50..6aea7b1 100644
--- a/docs/quick_start.md
+++ b/docs/quick_start.md
@@ -3,7 +3,7 @@
This is a quick start guide that will show you how to run the keyword spotting example application.
The aim of this quick start guide is to enable you to run an application quickly on the Fixed Virtual Platform.
The assumption we are making is that your Arm® Ethos™-U55 NPU is configured to use 128 Multiply-Accumulate units,
-is using a shared SRAM with the Arm® Cortex®-M55.
+is using a shared SRAM with the Arm® Cortex®-M55.
1. Verify you have installed [the required prerequisites](sections/building.md#Build-prerequisites).
@@ -58,11 +58,11 @@
--output ./resources_downloaded/ad/ifm0.npy
curl -L https://github.com/ARM-software/ML-zoo/raw/7c32b097f7d94aae2cd0b98a8ed5a3ba81e66b18/models/anomaly_detection/micronet_medium/tflite_int8/testing_output/Identity/0.npy \
--output ./resources_downloaded/ad/ofm0.npy
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/wav2letter_int8.tflite \
- --output ./resources_downloaded/asr/wav2letter_int8.tflite
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_input/input_2_int8/0.npy \
+curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/wav2letter_pruned_int8.tflite \
+ --output ./resources_downloaded/asr/wav2letter_pruned_int8.tflite
+curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_input/input_2_int8/0.npy \
--output ./resources_downloaded/asr/ifm0.npy
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_output/Identity_int8/0.npy \
+curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_output/Identity_int8/0.npy \
--output ./resources_downloaded/asr/ofm0.npy
curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/image_classification/mobilenet_v2_1.0_224/tflite_uint8/mobilenet_v2_1.0_224_quantized_1_default_1.tflite \
--output ./resources_downloaded/img_class/mobilenet_v2_1.0_224_quantized_1_default_1.tflite
@@ -76,13 +76,11 @@
--output ./resources_downloaded/kws/ifm0.npy
curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/testing_output/Identity/0.npy \
--output ./resources_downloaded/kws/ofm0.npy
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/wav2letter_int8.tflite \
- --output ./resources_downloaded/kws_asr/wav2letter_int8.tflite
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_input/input_2_int8/0.npy \
+curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/wav2letter_pruned_int8.tflite \
+ --output ./resources_downloaded/kws_asr/wav2letter_pruned_int8.tflite
+curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_input/input_2_int8/0.npy \
--output ./resources_downloaded/kws_asr/asr/ifm0.npy
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_input/input_2_int8/0.npy
- --output ./resources_downloaded/kws_asr/asr/ifm0.npy
-curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_output/Identity_int8/0.npy \
+curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_output/Identity_int8/0.npy \
--output ./resources_downloaded/kws_asr/asr/ofm0.npy
curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/ds_cnn_clustered_int8.tflite \
--output ./resources_downloaded/kws_asr/ds_cnn_clustered_int8.tflite
diff --git a/docs/sections/memory_considerations.md b/docs/sections/memory_considerations.md
index 48651f1..4727711 100644
--- a/docs/sections/memory_considerations.md
+++ b/docs/sections/memory_considerations.md
@@ -1,7 +1,5 @@
# Memory considerations
-## Contents
-
- [Memory considerations](#memory-considerations)
- [Introduction](#introduction)
- [Understanding memory usage from Vela output](#understanding-memory-usage-from-vela-output)
@@ -114,9 +112,9 @@
- 638.6 KiB of SRAM
- 3.1 MB of flash memory.
-- [Automated speech recognition](https://github.com/ARM-software/ML-zoo/tree/master/models/speech_recognition/wav2letter/tflite_int8) requires
- - 635.3 KiB of SRAM
- - 21.1 MB of flash memory.
+- [Automated speech recognition](https://github.com/ARM-software/ML-zoo/tree/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8) requires
+ - 655.16 KiB of SRAM
+ - 13.42 MB of flash memory.
## Memory constraints
diff --git a/set_up_default_resources.py b/set_up_default_resources.py
index 79b0333..7639364 100755
--- a/set_up_default_resources.py
+++ b/set_up_default_resources.py
@@ -36,12 +36,12 @@
},
{
"use_case_name": "asr",
- "resources": [{"name": "wav2letter_int8.tflite",
- "url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/wav2letter_int8.tflite"},
+ "resources": [{"name": "wav2letter_pruned_int8.tflite",
+ "url": "https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/wav2letter_pruned_int8.tflite"},
{"name": "ifm0.npy",
- "url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_input/input_2_int8/0.npy"},
+ "url": "https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_input/input_2_int8/0.npy"},
{"name": "ofm0.npy",
- "url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_output/Identity_int8/0.npy"}]
+ "url": "https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_output/Identity_int8/0.npy"}]
},
{
"use_case_name": "img_class",
@@ -63,12 +63,12 @@
},
{
"use_case_name": "kws_asr",
- "resources": [{"name": "wav2letter_int8.tflite",
- "url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/wav2letter_int8.tflite"},
+ "resources": [{"name": "wav2letter_pruned_int8.tflite",
+ "url": "https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/wav2letter_pruned_int8.tflite"},
{"sub_folder": "asr", "name": "ifm0.npy",
- "url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_input/input_2_int8/0.npy"},
+ "url": "https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_input/input_2_int8/0.npy"},
{"sub_folder": "asr", "name": "ofm0.npy",
- "url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/speech_recognition/wav2letter/tflite_int8/testing_output/Identity_int8/0.npy"},
+ "url": "https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_output/Identity_int8/0.npy"},
{"name": "ds_cnn_clustered_int8.tflite",
"url": "https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/ds_cnn_clustered_int8.tflite"},
{"sub_folder": "kws", "name": "ifm0.npy",
diff --git a/source/use_case/asr/usecase.cmake b/source/use_case/asr/usecase.cmake
index 3d18997..89014de 100644
--- a/source/use_case/asr/usecase.cmake
+++ b/source/use_case/asr/usecase.cmake
@@ -63,7 +63,7 @@
# Generate labels file
set(${use_case}_LABELS_CPP_FILE Labels)
generate_labels_code(
- INPUT "${${use_case}_LABELS_TXT_FILE}"
+ INPUT "${${use_case}_LABELS_TXT_FILE}"
DESTINATION_SRC ${SRC_GEN_DIR}
DESTINATION_HDR ${INC_GEN_DIR}
OUTPUT_FILENAME "${${use_case}_LABELS_CPP_FILE}"
@@ -75,9 +75,9 @@
STRING)
if (ETHOS_U55_ENABLED)
- set(DEFAULT_MODEL_PATH ${DEFAULT_MODEL_DIR}/wav2letter_int8_vela_H128.tflite)
+ set(DEFAULT_MODEL_PATH ${DEFAULT_MODEL_DIR}/wav2letter_pruned_int8_vela_H128.tflite)
else()
- set(DEFAULT_MODEL_PATH ${DEFAULT_MODEL_DIR}/wav2letter_int8.tflite)
+ set(DEFAULT_MODEL_PATH ${DEFAULT_MODEL_DIR}/wav2letter_pruned_int8.tflite)
endif()
set(EXTRA_MODEL_CODE
diff --git a/source/use_case/kws_asr/usecase.cmake b/source/use_case/kws_asr/usecase.cmake
index 9ff4fff..c47d633 100644
--- a/source/use_case/kws_asr/usecase.cmake
+++ b/source/use_case/kws_asr/usecase.cmake
@@ -68,10 +68,10 @@
if (ETHOS_U55_ENABLED)
set(DEFAULT_MODEL_PATH_KWS ${DEFAULT_MODEL_DIR}/ds_cnn_clustered_int8_vela_H128.tflite)
- set(DEFAULT_MODEL_PATH_ASR ${DEFAULT_MODEL_DIR}/wav2letter_int8_vela_H128.tflite)
+ set(DEFAULT_MODEL_PATH_ASR ${DEFAULT_MODEL_DIR}/wav2letter_pruned_int8_vela_H128.tflite)
else()
set(DEFAULT_MODEL_PATH_KWS ${DEFAULT_MODEL_DIR}/ds_cnn_clustered_int8.tflite)
- set(DEFAULT_MODEL_PATH_ASR ${DEFAULT_MODEL_DIR}/wav2letter_int8.tflite)
+ set(DEFAULT_MODEL_PATH_ASR ${DEFAULT_MODEL_DIR}/wav2letter_pruned_int8.tflite)
endif()
USER_OPTION(${use_case}_MODEL_TFLITE_PATH_KWS "NN models file to be used for KWS in the evaluation application. Model files must be in tflite format."