From 01293fe612422cbc142b56ff513825c35f532ba0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Marcin=20W=C4=85troba?= <markowanga@gmail.com>
Date: Sat, 14 Jan 2023 00:29:09 +0100
Subject: [PATCH] Change facebook wav2vec2 model
---
new_experiment/add_to_queue_pipeline.py | 31 ++++++++++---------
.../pipeline/pipeline_process_asr.py | 6 ++--
.../pipeline_process_spacy_dep_tag_wer.py | 6 +---
.../pipeline_process_spacy_ner_wer.py | 6 +---
.../pipeline_process_spacy_pos_wer.py | 3 +-
.../pipeline_process_wikineural_ner_wer.py | 1 -
.../pipeline_process_word_classic_wer.py | 2 +-
.../pipeline_process_word_embedding_wer.py | 8 +----
.../{new_worker.py => worker_asr.py} | 0
.../{worker.py => worker_pipeline.py} | 0
10 files changed, 24 insertions(+), 39 deletions(-)
rename new_experiment/{new_worker.py => worker_asr.py} (100%)
rename new_experiment/{worker.py => worker_pipeline.py} (100%)
diff --git a/new_experiment/add_to_queue_pipeline.py b/new_experiment/add_to_queue_pipeline.py
index b929cf0..9b9728b 100644
--- a/new_experiment/add_to_queue_pipeline.py
+++ b/new_experiment/add_to_queue_pipeline.py
@@ -49,30 +49,31 @@ def add_whisper(channel: BlockingChannel):
asr_name = f'whisper_{whisper_variant}'
for dataset_name in get_all_datasets_with_language():
for command in COMMANDS:
- add_to_queue(dataset_name, asr_name, command, channel)
+ add_to_queue(dataset_name, asr_name, command, channel, 'asr_benchmark_experiments')
def get_hf_facebook_wav2vec2_model_by_language_code(language_code: str) -> str:
return {
- 'nl': 'facebook_wav2vec2_large_xlsr_53_dutch',
- 'en': 'facebook_wav2vec2_xls_r_300m',
- 'fr': 'facebook_wav2vec2_large_xlsr_53_french',
- 'de': 'facebook_wav2vec2_large_xlsr_53_german',
- 'it': 'facebook_wav2vec2_large_xlsr_53_italian',
- 'pl': 'facebook_wav2vec2_large_xlsr_53_polish',
- 'es': 'facebook_wav2vec2_large_xlsr_53_spanish'
+ # 'nl': 'facebook_wav2vec2_large_xlsr_53_dutch',
+ 'en': 'facebook_wav2vec2_large_960h_lv60_self',
+ # 'fr': 'facebook_wav2vec2_large_xlsr_53_french',
+ # 'de': 'facebook_wav2vec2_large_xlsr_53_german',
+ # 'it': 'facebook_wav2vec2_large_xlsr_53_italian',
+ # 'pl': 'facebook_wav2vec2_large_xlsr_53_polish',
+ # 'es': 'facebook_wav2vec2_large_xlsr_53_spanish'
}[language_code]
def add_facebook_hf_wav2vec2_asr(channel: BlockingChannel):
for dataset_name in get_all_datasets_with_language():
- add_to_queue(
- dataset_name,
- get_hf_facebook_wav2vec2_model_by_language_code(dataset_name[:2]),
- 'hf_facebook_wav2vec2_asr',
- channel,
- 'hf_facebook_wav2vec2_asr'
- )
+ if dataset_name.startswith('en'):
+ add_to_queue(
+ dataset_name,
+ get_hf_facebook_wav2vec2_model_by_language_code(dataset_name[:2]),
+ 'hf_facebook_wav2vec2_asr',
+ channel,
+ 'hf_facebook_wav2vec2_asr'
+ )
def main():
diff --git a/new_experiment/pipeline/pipeline_process_asr.py b/new_experiment/pipeline/pipeline_process_asr.py
index c883a3d..1860df6 100644
--- a/new_experiment/pipeline/pipeline_process_asr.py
+++ b/new_experiment/pipeline/pipeline_process_asr.py
@@ -10,8 +10,8 @@ from sziszapangma.integration.task.asr_task import AsrTask
def get_asr_processor(asr_name: str) -> AsrProcessor:
if asr_name == 'facebook_wav2vec2_large_xlsr_53_dutch':
return Wav2Vec2AsrProcessor('facebook/wav2vec2-large-xlsr-53-dutch')
- if asr_name == 'facebook_wav2vec2_xls_r_300m':
- return Wav2Vec2AsrProcessor('facebook/wav2vec2-xls-r-300m')
+ if asr_name == 'facebook_wav2vec2_large_960h_lv60_self':
+ return Wav2Vec2AsrProcessor('facebook/wav2vec2-large-960h-lv60-self')
if asr_name == 'facebook_wav2vec2_large_xlsr_53_french':
return Wav2Vec2AsrProcessor('facebook/wav2vec2-large-xlsr-53-french')
if asr_name == 'facebook_wav2vec2_large_xlsr_53_german':
@@ -34,7 +34,7 @@ def run_hf_facebook_wav2vec2_asr_task(dataset_name: str, asr_name: str):
AsrTask(
asr_property_name=PropertyHelper.asr_result(asr_name),
task_name=f'AsrTask___{dataset_name}___{asr_name}',
- require_update=True,
+ require_update=False,
asr_processor=get_asr_processor(asr_name),
record_path_provider=record_provider
)
diff --git a/new_experiment/pipeline/pipeline_process_spacy_dep_tag_wer.py b/new_experiment/pipeline/pipeline_process_spacy_dep_tag_wer.py
index 8e2d568..c66a8f3 100644
--- a/new_experiment/pipeline/pipeline_process_spacy_dep_tag_wer.py
+++ b/new_experiment/pipeline/pipeline_process_spacy_dep_tag_wer.py
@@ -1,7 +1,3 @@
-import argparse
-
-from experiment.const_pipeline_names import GOLD_TRANSCRIPT
-from experiment.experiment_dependency_provider import get_record_provider, get_repository
from experiment.sentence_wer_processor.spacy_pos_sentence_dep_tag_processor import SpacyDepTagSentenceWerProcessor
from new_experiment.new_dependency_provider import get_experiment_repository, get_minio_audio_record_repository
from new_experiment.utils.get_spacy_model_name import get_spacy_model_name
@@ -25,7 +21,7 @@ def run_spacy_dep_tag_wer_pipeline(dataset_name: str, asr_name: str):
alignment_property_name=PropertyHelper.dep_tag_alignment(asr_name, model_name),
wer_property_name=PropertyHelper.dep_tag_metrics(asr_name, model_name),
task_name=f'SpacyDepTagSentenceWerProcessor___{dataset_name}___{asr_name}',
- require_update=True
+ require_update=False
)
],
experiment_repository=repository,
diff --git a/new_experiment/pipeline/pipeline_process_spacy_ner_wer.py b/new_experiment/pipeline/pipeline_process_spacy_ner_wer.py
index c74e419..9cdf4b0 100644
--- a/new_experiment/pipeline/pipeline_process_spacy_ner_wer.py
+++ b/new_experiment/pipeline/pipeline_process_spacy_ner_wer.py
@@ -1,7 +1,3 @@
-import argparse
-
-from experiment.const_pipeline_names import GOLD_TRANSCRIPT
-from experiment.experiment_dependency_provider import get_record_provider, get_repository
from experiment.sentence_wer_processor.spacy_ner_sentence_wer_processor import SpacyNerSentenceWerProcessor
from new_experiment.new_dependency_provider import get_experiment_repository, get_minio_audio_record_repository
from new_experiment.utils.get_spacy_model_name import get_spacy_model_name
@@ -25,7 +21,7 @@ def run_spacy_ner_wer_pipeline(dataset_name: str, asr_name: str):
alignment_property_name=PropertyHelper.ner_alignment(asr_name, model_name),
wer_property_name=PropertyHelper.ner_metrics(asr_name, model_name),
task_name=f'SpacyNerSentenceWerProcessor___{dataset_name}___{asr_name}',
- require_update=True
+ require_update=False
)
],
experiment_repository=repository,
diff --git a/new_experiment/pipeline/pipeline_process_spacy_pos_wer.py b/new_experiment/pipeline/pipeline_process_spacy_pos_wer.py
index 5959dc9..1af0b65 100644
--- a/new_experiment/pipeline/pipeline_process_spacy_pos_wer.py
+++ b/new_experiment/pipeline/pipeline_process_spacy_pos_wer.py
@@ -1,4 +1,3 @@
-from experiment.experiment_dependency_provider import get_record_provider, get_repository
from experiment.sentence_wer_processor.spacy_pos_sentence_wer_processor import SpacyPosSentenceWerProcessor
from new_experiment.new_dependency_provider import get_experiment_repository, get_minio_audio_record_repository
from new_experiment.utils.get_spacy_model_name import get_spacy_model_name
@@ -22,7 +21,7 @@ def run_spacy_pos_wer_pipeline(dataset_name: str, asr_name: str):
alignment_property_name=PropertyHelper.pos_alignment(asr_name, model_name),
wer_property_name=PropertyHelper.pos_metrics(asr_name, model_name),
task_name=f'SpacyPosSentenceWerProcessor___{dataset_name}___{asr_name}',
- require_update=True
+ require_update=False
)
],
experiment_repository=repository,
diff --git a/new_experiment/pipeline/pipeline_process_wikineural_ner_wer.py b/new_experiment/pipeline/pipeline_process_wikineural_ner_wer.py
index 761abb4..3675fc9 100644
--- a/new_experiment/pipeline/pipeline_process_wikineural_ner_wer.py
+++ b/new_experiment/pipeline/pipeline_process_wikineural_ner_wer.py
@@ -1,4 +1,3 @@
-from experiment.experiment_dependency_provider import get_record_provider, get_repository
from experiment.sentence_wer_processor.wikineural_multilingual_ner_transformers_wer_processor_base import \
WikineuralMultilingualNerTransformersWerProcessorBase
from new_experiment.new_dependency_provider import get_experiment_repository, get_minio_audio_record_repository
diff --git a/new_experiment/pipeline/pipeline_process_word_classic_wer.py b/new_experiment/pipeline/pipeline_process_word_classic_wer.py
index 231a696..212e3db 100644
--- a/new_experiment/pipeline/pipeline_process_word_classic_wer.py
+++ b/new_experiment/pipeline/pipeline_process_word_classic_wer.py
@@ -15,7 +15,7 @@ def run_word_wer_classic_pipeline(dataset_name: str, asr_name: str):
asr_property_name=PropertyHelper.asr_result(asr_name),
gold_transcript_property_name=PropertyHelper.get_gold_transcript_words(),
metrics_property_name=PropertyHelper.word_wer_classic_metrics(asr_name),
- require_update=True,
+ require_update=False,
alignment_property_name=PropertyHelper.word_wer_classic_alignment(asr_name)
),
],
diff --git a/new_experiment/pipeline/pipeline_process_word_embedding_wer.py b/new_experiment/pipeline/pipeline_process_word_embedding_wer.py
index 8f942f0..9026787 100644
--- a/new_experiment/pipeline/pipeline_process_word_embedding_wer.py
+++ b/new_experiment/pipeline/pipeline_process_word_embedding_wer.py
@@ -1,14 +1,8 @@
-import argparse
-
-from experiment.const_pipeline_names import GOLD_TRANSCRIPT
-from experiment.experiment_dependency_provider import get_record_provider, get_repository
from new_experiment.new_dependency_provider import get_experiment_repository, get_minio_audio_record_repository
from new_experiment.utils.loaded_remote_dataset_helper import LoadedRemoteDatasetHelper
from new_experiment.utils.property_helper import PropertyHelper
from sziszapangma.core.transformer.fasttext_embedding_transformer import FasttextEmbeddingTransformer
-from sziszapangma.core.transformer.web_embedding_transformer import WebEmbeddingTransformer
from sziszapangma.integration.experiment_manager import ExperimentManager
-from sziszapangma.integration.task.classic_wer_metric_task import ClassicWerMetricTask
from sziszapangma.integration.task.embedding_wer_metrics_task import EmbeddingWerMetricsTask
@@ -22,7 +16,7 @@ def run_word_wer_embedding_pipeline(dataset_name: str, asr_name: str):
asr_property_name=PropertyHelper.asr_result(asr_name),
gold_transcript_property_name=PropertyHelper.get_gold_transcript_words(),
metrics_property_name=PropertyHelper.word_wer_embeddings_metrics(asr_name),
- require_update=True,
+ require_update=False,
embedding_transformer=FasttextEmbeddingTransformer(dataset_name[:2]),
alignment_property_name=PropertyHelper.word_wer_embeddings_alignment(asr_name)
)
diff --git a/new_experiment/new_worker.py b/new_experiment/worker_asr.py
similarity index 100%
rename from new_experiment/new_worker.py
rename to new_experiment/worker_asr.py
diff --git a/new_experiment/worker.py b/new_experiment/worker_pipeline.py
similarity index 100%
rename from new_experiment/worker.py
rename to new_experiment/worker_pipeline.py
--
GitLab