experiments/STT
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge pull request #2984 from tilmankamp/cv-train-all

Browse Source 
CV2 importer: train-all.csv
This commit is contained in:
Tilman Kamp 2020-05-07 16:54:44 +02:00 committed by GitHub
commit ef81bb9512
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 54 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
bin/import_cv2.py
View File

@ -31,11 +31,14 @@ MAX_SECS = 10
def _preprocess_data(tsv_dir, audio_dir, space_after_every_character=False): def _preprocess_data(tsv_dir, audio_dir, space_after_every_character=False):
for dataset in ["train", "test", "dev", "validated", "other"]: exclude = []
input_tsv = os.path.join(os.path.abspath(tsv_dir), dataset + ".tsv") for dataset in ["test", "dev", "train", "validated", "other"]:
if os.path.isfile(input_tsv): set_samples = _maybe_convert_set(dataset, tsv_dir, audio_dir, space_after_every_character)
print("Loading TSV file: ", input_tsv) if dataset in ["test", "dev"]:
_maybe_convert_set(input_tsv, audio_dir, space_after_every_character) exclude += set_samples
if dataset == "validated":
_maybe_convert_set("train-all", tsv_dir, audio_dir, space_after_every_character,
rows=set_samples, exclude=exclude)
def one_sample(sample): def one_sample(sample):
@ -72,29 +75,36 @@ def one_sample(sample):
counter["too_long"] += 1 counter["too_long"] += 1
else: else:
# This one is good - keep it for the target CSV # This one is good - keep it for the target CSV
rows.append((os.path.split(wav_filename)[-1], file_size, label)) rows.append((os.path.split(wav_filename)[-1], file_size, label, sample[2]))
counter["all"] += 1 counter["all"] += 1
counter["total_time"] += frames counter["total_time"] += frames
return (counter, rows) return (counter, rows)
def _maybe_convert_set(input_tsv, audio_dir, space_after_every_character=None): def _maybe_convert_set(dataset, tsv_dir, audio_dir, space_after_every_character=None, rows=None, exclude=None):
output_csv = os.path.join( exclude_transcripts = set()
audio_dir, os.path.split(input_tsv)[-1].replace("tsv", "csv") exclude_speakers = set()
) if exclude is not None:
print("Saving new DeepSpeech-formatted CSV file to: ", output_csv) for sample in exclude:
exclude_transcripts.add(sample[2])
exclude_speakers.add(sample[3])
if rows is None:
rows = []
input_tsv = os.path.join(os.path.abspath(tsv_dir), dataset + ".tsv")
if not os.path.isfile(input_tsv):
return rows
print("Loading TSV file: ", input_tsv)
# Get audiofile path and transcript for each sentence in tsv # Get audiofile path and transcript for each sentence in tsv
samples = [] samples = []
with open(input_tsv, encoding="utf-8") as input_tsv_file: with open(input_tsv, encoding="utf-8") as input_tsv_file:
reader = csv.DictReader(input_tsv_file, delimiter="\t") reader = csv.DictReader(input_tsv_file, delimiter="\t")
for row in reader: for row in reader:
samples.append((os.path.join(audio_dir, row["path"]), row["sentence"])) samples.append((os.path.join(audio_dir, row["path"]), row["sentence"], row["client_id"]))
counter = get_counter() counter = get_counter()
num_samples = len(samples) num_samples = len(samples)
rows = []
print("Importing mp3 files...") print("Importing mp3 files...")
pool = Pool() pool = Pool()
@ -107,12 +117,21 @@ def _maybe_convert_set(input_tsv, audio_dir, space_after_every_character=None):
pool.close() pool.close()
pool.join() pool.join()
imported_samples = get_imported_samples(counter)
assert counter["all"] == num_samples
assert len(rows) == imported_samples
print_import_report(counter, SAMPLE_RATE, MAX_SECS)
output_csv = os.path.join(os.path.abspath(audio_dir), dataset + ".csv")
print("Saving new DeepSpeech-formatted CSV file to: ", output_csv)
with open(output_csv, "w", encoding="utf-8") as output_csv_file: with open(output_csv, "w", encoding="utf-8") as output_csv_file:
print("Writing CSV file for DeepSpeech.py as: ", output_csv) print("Writing CSV file for DeepSpeech.py as: ", output_csv)
writer = csv.DictWriter(output_csv_file, fieldnames=FIELDNAMES) writer = csv.DictWriter(output_csv_file, fieldnames=FIELDNAMES)
writer.writeheader() writer.writeheader()
bar = progressbar.ProgressBar(max_value=len(rows), widgets=SIMPLE_BAR) bar = progressbar.ProgressBar(max_value=len(rows), widgets=SIMPLE_BAR)
for filename, file_size, transcript in bar(rows): for filename, file_size, transcript, speaker in bar(rows):
if transcript in exclude_transcripts or speaker in exclude_speakers:
continue
if space_after_every_character: if space_after_every_character:
writer.writerow( writer.writerow(
{ {
@ -129,12 +148,7 @@ def _maybe_convert_set(input_tsv, audio_dir, space_after_every_character=None):
"transcript": transcript, "transcript": transcript,
} }
) )
return rows
imported_samples = get_imported_samples(counter)
assert counter["all"] == num_samples
assert len(rows) == imported_samples
print_import_report(counter, SAMPLE_RATE, MAX_SECS)
def _maybe_convert_wav(mp3_filename, wav_filename): def _maybe_convert_wav(mp3_filename, wav_filename):