Add WhisperX model download script and configuration files for version 3.4.2

tools/whisperx/download_models_3.4.2.py

@@ -0,0 +1,168 @@
+import argparse
+import os
+import urllib.request
+from pathlib import Path
+
+import huggingface_hub
+
+DEFAULT_ALIGN_MODELS_TORCH = {
+    "en": "WAV2VEC2_ASR_BASE_960H",
+    "fr": "VOXPOPULI_ASR_BASE_10K_FR",
+    "de": "VOXPOPULI_ASR_BASE_10K_DE",
+    "es": "VOXPOPULI_ASR_BASE_10K_ES",
+    "it": "VOXPOPULI_ASR_BASE_10K_IT",
+}
+
+DEFAULT_ALIGN_MODELS_HF = {
+    "ja": "jonatasgrosman/wav2vec2-large-xlsr-53-japanese",
+    "zh": "jonatasgrosman/wav2vec2-large-xlsr-53-chinese-zh-cn",
+    "nl": "jonatasgrosman/wav2vec2-large-xlsr-53-dutch",
+    "uk": "Yehor/wav2vec2-xls-r-300m-uk-with-small-lm",
+    "pt": "jonatasgrosman/wav2vec2-large-xlsr-53-portuguese",
+    "ar": "jonatasgrosman/wav2vec2-large-xlsr-53-arabic",
+    "cs": "comodoro/wav2vec2-xls-r-300m-cs-250",
+    "ru": "jonatasgrosman/wav2vec2-large-xlsr-53-russian",
+    "pl": "jonatasgrosman/wav2vec2-large-xlsr-53-polish",
+    "hu": "jonatasgrosman/wav2vec2-large-xlsr-53-hungarian",
+    "fi": "jonatasgrosman/wav2vec2-large-xlsr-53-finnish",
+    "fa": "jonatasgrosman/wav2vec2-large-xlsr-53-persian",
+    "el": "jonatasgrosman/wav2vec2-large-xlsr-53-greek",
+    "tr": "mpoyraz/wav2vec2-xls-r-300m-cv7-turkish",
+    "da": "saattrupdan/wav2vec2-xls-r-300m-ftspeech",
+    "he": "imvladikon/wav2vec2-xls-r-300m-hebrew",
+    "vi": "nguyenvulebinh/wav2vec2-base-vi",
+    "ko": "kresnik/wav2vec2-large-xlsr-korean",
+    "ur": "kingabzpro/wav2vec2-large-xls-r-300m-Urdu",
+    "te": "anuragshas/wav2vec2-large-xlsr-53-telugu",
+    "hi": "theainerd/Wav2Vec2-large-xlsr-hindi",
+    "ca": "softcatala/wav2vec2-large-xlsr-catala",
+    "ml": "gvs/wav2vec2-large-xlsr-malayalam",
+    "no": "NbAiLab/nb-wav2vec2-1b-bokmaal-v2",
+    "nn": "NbAiLab/nb-wav2vec2-1b-nynorsk",
+    "sk": "comodoro/wav2vec2-xls-r-300m-sk-cv8",
+    "sl": "anton-l/wav2vec2-large-xlsr-53-slovenian",
+    "hr": "classla/wav2vec2-xls-r-parlaspeech-hr",
+    "ro": "gigant/romanian-wav2vec2",
+    "eu": "stefan-it/wav2vec2-large-xlsr-53-basque",
+    "gl": "ifrz/wav2vec2-large-xlsr-galician",
+    "ka": "xsway/wav2vec2-large-xlsr-georgian",
+    "lv": "jimregan/wav2vec2-large-xlsr-latvian-cv",
+    "tl": "Khalsuu/filipino-wav2vec2-l-xls-r-300m-official",
+    "sv": "KBLab/wav2vec2-large-voxrex-swedish",
+}
+
+_MODELS = {
+    "tiny": "Systran/faster-whisper-tiny",
+    "base": "Systran/faster-whisper-base",
+    "small": "Systran/faster-whisper-small",
+    "medium": "Systran/faster-whisper-medium",
+    "large": "Systran/faster-whisper-large-v3",
+    "turbo": "mobiuslabsgmbh/faster-whisper-large-v3-turbo",
+}
+
+TORCHAUDIO_MODEL_FILES = {
+    "WAV2VEC2_ASR_BASE_960H": "wav2vec2_fairseq_base_ls960_asr_ls960.pth",
+    "VOXPOPULI_ASR_BASE_10K_FR": "wav2vec2_voxpopuli_base_10k_asr_fr.pt",
+    "VOXPOPULI_ASR_BASE_10K_DE": "wav2vec2_voxpopuli_base_10k_asr_de.pt",
+    "VOXPOPULI_ASR_BASE_10K_ES": "wav2vec2_voxpopuli_base_10k_asr_es.pt",
+    "VOXPOPULI_ASR_BASE_10K_IT": "wav2vec2_voxpopuli_base_10k_asr_it.pt",
+}
+
+TORCHAUDIO_BASE_URL = "https://download.pytorch.org/torchaudio/models/"
+
+
+def download_torchaudio_models(model_dir):
+    checkpoints_dir = Path(model_dir) / "hub" / "checkpoints"
+    checkpoints_dir.mkdir(parents=True, exist_ok=True)
+    for pipeline_name, filename in TORCHAUDIO_MODEL_FILES.items():
+        dest = checkpoints_dir / filename
+        if dest.exists():
+            print(f"Already exists, skipping: {filename}")
+            continue
+        url = TORCHAUDIO_BASE_URL + filename
+        print(f"Downloading {pipeline_name} from {url} ...")
+        try:
+            urllib.request.urlretrieve(url, dest)
+            print(f"Downloaded: {filename}")
+        except Exception as e:
+            print(f"Failed to download {pipeline_name}: {e}")
+
+
+def download_hf_alignment_models(model_dir, token):
+    cache_dir = str(Path(model_dir) / "hub")
+    for lang, repo_id in DEFAULT_ALIGN_MODELS_HF.items():
+        print(f"Downloading HF alignment model [{lang}]: {repo_id} ...")
+        try:
+            huggingface_hub.snapshot_download(repo_id, cache_dir=cache_dir, token=token)
+            print(f"Downloaded: {repo_id}")
+        except Exception as e:
+            print(f"Failed to download {repo_id}: {e}")
+
+
+def download_asr_models(model_dir, token):
+    cache_dir = str(Path(model_dir) / "hub")
+    for model_name, repo_id in _MODELS.items():
+        print(f"Downloading ASR model [{model_name}]: {repo_id} ...")
+        try:
+            huggingface_hub.snapshot_download(
+                repo_id,
+                cache_dir=cache_dir,
+                token=token,
+                allow_patterns=[
+                    "config.json",
+                    "preprocessor_config.json",
+                    "model.bin",
+                    "tokenizer.json",
+                    "vocabulary.*",
+                ],
+            )
+            print(f"Downloaded: {model_name}")
+        except Exception as e:
+            print(f"Failed to download {model_name}: {e}")
+
+
+PYANNOTE_REPOS = [
+    "pyannote/speaker-diarization-3.1",
+    "pyannote/segmentation-3.0",
+    "pyannote/wespeaker-voxceleb-resnet34-LM",
+]
+
+
+def download_pyannote_model(model_dir, token):
+    cache_dir = str(Path(model_dir) / "hub")
+    for repo_id in PYANNOTE_REPOS:
+        print(f"Downloading {repo_id} ...")
+        try:
+            huggingface_hub.snapshot_download(repo_id, cache_dir=cache_dir, token=token)
+            print(f"Downloaded: {repo_id}")
+        except Exception as e:
+            print(f"Failed to download {repo_id}: {e}")
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Download WhisperX models (lightweight, no torch required)"
+    )
+    parser.add_argument(
+        "--model-dir",
+        default="whisperx_models",
+        help="Directory to store downloaded models",
+    )
+    parser.add_argument(
+        "--hf-token",
+        default=os.getenv("HF_AUTH_TOKEN"),
+        help="HuggingFace token for gated models",
+    )
+    args = parser.parse_args()
+
+    model_dir = args.model_dir
+    token = args.hf_token
+
+    download_torchaudio_models(model_dir)
+    download_hf_alignment_models(model_dir, token)
+    download_asr_models(model_dir, token)
+    download_pyannote_model(model_dir, token)
+
+
+if __name__ == "__main__":
+    main()

tools/whisperx/tool-data/huggingface.loc.sample

@@ -0,0 +1,78 @@
+# WhisperX model table — covers ASR and alignment models.
+#
+# This file uses the shared huggingface data table schema.
+# The pipeline_tag column distinguishes model types: asr | alignment.
+# The free_tag column is set to "whisperx" for all rows so the XML can
+# filter specifically for WhisperX rows in the shared table.
+#
+# To configure:
+# 1. Copy this file to huggingface.loc (removing .sample), or append to an
+#    existing huggingface.loc shared with other tools (e.g. Flux)
+# 2. Replace /data/db/whisperx_models with the actual root directory
+#    created by download_models_3.4.2.py
+# 3. Remove rows for models you have not downloaded
+#
+# Columns:
+#   value        : unique row identifier
+#   name         : display label shown in Galaxy UI
+#   pipeline_tag : asr | alignment
+#   domain       : whisper size key (asr) or BCP-47 language code (alignment)
+#   free_tag     : whisperx (used by the tool XML to filter rows)
+#   version      : tool version this row belongs to
+#   path         : root model directory (containing hub/, nltk_data/, etc.)
+#
+# Columns: value <TAB> name <TAB> pipeline_tag <TAB> domain <TAB> free_tag <TAB> version <TAB> path
+#
+# --- version 3.4.2 ---
+#
+# ── ASR models ──────────────────────────────────────────────────────────────────────
+whisperx_3.4.2_asr_turbo	Turbo (~8x faster than large)	asr	turbo	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_asr_tiny	Tiny (~10x faster than large)	asr	tiny	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_asr_base	Base (~7x faster than large)	asr	base	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_asr_small	Small (~4x faster than large)	asr	small	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_asr_medium	Medium (~2x faster than large)	asr	medium	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_asr_large	Large	asr	large	whisperx	3.4.2	/data/db/whisperx_models
+#
+# ── Alignment models — torchaudio (en, fr, de, es, it) ──────────────────────────────
+whisperx_3.4.2_align_en	English	alignment	en	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_fr	French	alignment	fr	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_de	German	alignment	de	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_es	Spanish	alignment	es	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_it	Italian	alignment	it	whisperx	3.4.2	/data/db/whisperx_models
+#
+# ── Alignment models — HuggingFace ──────────────────────────────────────────────────────
+whisperx_3.4.2_align_ja	Japanese	alignment	ja	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_zh	Chinese	alignment	zh	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_nl	Dutch	alignment	nl	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_uk	Ukrainian	alignment	uk	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_pt	Portuguese	alignment	pt	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ar	Arabic	alignment	ar	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_cs	Czech	alignment	cs	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ru	Russian	alignment	ru	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_pl	Polish	alignment	pl	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_hu	Hungarian	alignment	hu	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_fi	Finnish	alignment	fi	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_fa	Persian	alignment	fa	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_el	Greek	alignment	el	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_tr	Turkish	alignment	tr	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_da	Danish	alignment	da	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_he	Hebrew	alignment	he	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_vi	Vietnamese	alignment	vi	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ko	Korean	alignment	ko	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ur	Urdu	alignment	ur	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_te	Telugu	alignment	te	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_hi	Hindi	alignment	hi	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ca	Catalan	alignment	ca	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ml	Malayalam	alignment	ml	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_no	Norwegian	alignment	no	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_nn	Nynorsk	alignment	nn	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_sk	Slovak	alignment	sk	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_sl	Slovenian	alignment	sl	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_hr	Croatian	alignment	hr	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ro	Romanian	alignment	ro	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_eu	Basque	alignment	eu	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_gl	Galician	alignment	gl	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_ka	Georgian	alignment	ka	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_lv	Latvian	alignment	lv	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_tl	Tagalog	alignment	tl	whisperx	3.4.2	/data/db/whisperx_models
+whisperx_3.4.2_align_sv	Swedish	alignment	sv	whisperx	3.4.2	/data/db/whisperx_models

tools/whisperx/tool_data_table_conf.xml.sample

@@ -0,0 +1,13 @@
+<!-- Paths are relative to the value of `tool_data_path` in galaxy.ini -->
+<!-- WhisperX uses the shared huggingface data table.
+     ASR rows have pipeline_tag=asr; alignment rows have pipeline_tag=alignment.
+     Column layout (7 columns):
+       value, name, pipeline_tag, domain, free_tag, version, path
+     domain is the whisper size key (asr) or BCP-47 language code (alignment).
+     path is the root model directory (containing hub/, nltk_data/, etc.). -->
+<tables>
+    <table name="huggingface" comment_char="#" allow_duplicate_entries="False">
+        <columns>value, name, pipeline_tag, domain, free_tag, version, path</columns>
+        <file path="tool-data/huggingface.loc" />
+    </table>
+</tables>