batchalign 0.7.18.post12__tar.gz → 0.7.19__tar.gz

{batchalign-0.7.18.post12/batchalign.egg-info → batchalign-0.7.19}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: batchalign
-Version: 0.7.18.post12
+Version: 0.7.19
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -34,6 +34,7 @@ Requires-Dist: tiktoken
 Requires-Dist: blobfile
 Requires-Dist: sentencepiece
 Requires-Dist: googletrans
+Requires-Dist: openai-whisper>=20240930
 Provides-Extra: dev
 Requires-Dist: pytest; extra == "dev"
 Provides-Extra: train

{batchalign-0.7.18.post12 → batchalign-0.7.19}/batchalign/cli/cli.py

@@ -144,8 +144,10 @@ def align(ctx, in_dir, out_dir, whisper, wav2vec, **kwargs):
 
 @batchalign.command()
 @common_options
+@click.option("--whisper_oai/--rev",
+              default=False, help="Use the OpenAI's Whisper implementation instead of Rev.AI (default).")
 @click.option("--whisper/--rev",
-              default=False, help="Use OpenAI Whisper (ASR) instead of Rev.AI (default).")
+              default=False, help="Use Huggingface's Whisper implementation instead of Rev.AI (default).")
 @click.option("--whisperx/--rev",
               default=False, help="Use WhisperX instead of Rev.AI (default). Superceeds --whisper.")
 @click.option("--diarize/--nodiarize",
@@ -169,6 +171,8 @@ def transcribe(ctx, in_dir, out_dir, lang, num_speakers, **kwargs):
         asr = "whisper"
     if kwargs["whisperx"]:
         asr = "whisperx"
+    if kwargs["whisper_oai"]:
+        asr = "whisper_oai"
 
 
     def writer(doc, output):

{batchalign-0.7.18.post12 → batchalign-0.7.19}/batchalign/pipelines/__init__.py

@@ -1,6 +1,6 @@
 from .pipeline import BatchalignPipeline
 from .base import BatchalignEngine
-from .asr import WhisperEngine, RevEngine, WhisperXEngine
+from .asr import WhisperEngine, RevEngine, WhisperXEngine, OAIWhisperEngine
 
 from .morphosyntax import StanzaEngine, CorefEngine
 from .cleanup import NgramRetraceEngine, DisfluencyReplacementEngine

{batchalign-0.7.18.post12 → batchalign-0.7.19}/batchalign/pipelines/asr/__init__.py

@@ -1,3 +1,4 @@
 from .whisper import WhisperEngine
 from .rev import RevEngine
 from .whisperx import WhisperXEngine
+from .oai_whisper import OAIWhisperEngine

batchalign-0.7.19/batchalign/pipelines/asr/oai_whisper.py

@@ -0,0 +1,97 @@
+from batchalign.document import *
+from batchalign.pipelines.base import *
+from batchalign.pipelines.asr.utils import *
+from batchalign.models import WhisperASRModel, BertUtteranceModel, BertCantoneseUtteranceModel
+
+import pycountry
+
+import logging
+L = logging.getLogger("batchalign")
+
+from batchalign.utils.utils import correct_timing
+from batchalign.models import resolve
+
+import whisper
+
+class OAIWhisperEngine(BatchalignEngine):
+
+    @property
+    def tasks(self):
+        # if there is no utterance segmentation scheme, we only
+        # run ASR
+        if self.__engine:
+            return [ Task.ASR, Task.UTTERANCE_SEGMENTATION ]
+        else:
+            return [ Task.ASR ]
+
+    def __init__(self, model=None, lang="eng"):
+
+        # try to resolve our internal model
+        res = resolve("whisper", lang)
+        if res:
+            model, base = res
+        else:
+            model = "openai/whisper-large-v3"
+            base = "openai/whisper-large-v3"
+
+        if lang == "mys":
+            language = "Malay"
+        else:
+            language = pycountry.languages.get(alpha_3=lang).name
+        if language == "Yue Chinese":
+            language = "Cantonese"
+        if "greek" in language.lower():
+            language = "Greek"
+
+            
+        self.__whisper = whisper.load_model("turbo")
+        self.__lang = lang
+        self.__language = language
+
+        if resolve("utterance", self.__lang) != None:
+            L.debug("Initializing utterance model...")
+            if lang != "yue":
+                self.__engine = BertUtteranceModel(resolve("utterance", lang))
+            else:
+                # we have special inference procedure for cantonese
+                self.__engine = BertCantoneseUtteranceModel(resolve("utterance", lang))
+            L.debug("Done.")
+        else:
+            self.__engine = None
+
+    def generate(self, source_path, **kwargs):
+        res = self.__whisper.transcribe(source_path,
+                                        word_timestamps=True,
+                                        language=self.__language)
+        turns = []
+        for i in res["segments"]:
+            turn = []
+            for j in i["words"]:
+                turn.append({
+                    "type": "text",
+                    "ts": j["start"],
+                    "end_ts": j["end"],
+                    "value": j["word"]
+                })
+            turns.append({
+                "elements": turn,
+                "speaker": 0
+            })
+        doc = process_generation({"monologues": turns},
+                                    self.__lang,
+                                    utterance_engine=self.__engine)
+        # define media tier
+        media = Media(type=MediaType.AUDIO, name=Path(source_path).stem, url=source_path)
+        doc.media = media
+
+        return correct_timing(doc)
+
+
+    # model="openai/whisper-large-v2", language="english"
+
+# e = WhisperEngine()
+# tmp = e.generate("./batchalign/tests/pipelines/asr/support/test.mp3", 1)
+# tmp.model_dump()
+# file = "./batchalign/tests/pipelines/asr/support/test.mp3"
+
+

{batchalign-0.7.18.post12 → batchalign-0.7.19}/batchalign/pipelines/dispatch.py

@@ -7,7 +7,7 @@ from batchalign import (WhisperEngine, WhisperFAEngine, StanzaEngine, RevEngine,
                         NgramRetraceEngine, DisfluencyReplacementEngine, WhisperUTREngine,
                         RevUTREngine, EvaluationEngine, WhisperXEngine, NemoSpeakerEngine,
                         StanzaUtteranceEngine, CorefEngine, Wave2VecFAEngine, SeamlessTranslationModel,
-                        GoogleTranslateEngine)
+                        GoogleTranslateEngine, OAIWhisperEngine)
 from batchalign import BatchalignPipeline
 from batchalign.models import resolve
 
@@ -19,7 +19,7 @@ L = logging.getLogger("batchalign")
 
 # default for all languages
 DEFAULT_PACKAGES = {
-    "asr": "whisper",
+    "asr": "whisper_oai",
     "utr": "whisper_utr",
     "fa": "whisper_fa",
     "speaker": "nemo_speaker",
@@ -135,7 +135,8 @@ def dispatch_pipeline(pkg_str, lang, num_speakers=None, **arg_overrides):
             engines.append(SeamlessTranslationModel())
         elif engine == "gtrans":
             engines.append(GoogleTranslateEngine())
-
+        elif engine == "whisper_oai":
+            engines.append(OAIWhisperEngine())
 
     L.debug(f"Done initalizing packages.")
     return BatchalignPipeline(*engines)

batchalign-0.7.19/batchalign/version

@@ -0,0 +1,3 @@
+0.7.19
+May 9th, 2025
+OpenAI Whisper Implementation

{batchalign-0.7.18.post12 → batchalign-0.7.19/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: batchalign
-Version: 0.7.18.post12
+Version: 0.7.19
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -34,6 +34,7 @@ Requires-Dist: tiktoken
 Requires-Dist: blobfile
 Requires-Dist: sentencepiece
 Requires-Dist: googletrans
+Requires-Dist: openai-whisper>=20240930
 Provides-Extra: dev
 Requires-Dist: pytest; extra == "dev"
 Provides-Extra: train

{batchalign-0.7.18.post12 → batchalign-0.7.19}/batchalign.egg-info/SOURCES.txt

@@ -59,6 +59,7 @@ batchalign/pipelines/analysis/__init__.py
 batchalign/pipelines/analysis/eval.py
 batchalign/pipelines/asr/__init__.py
 batchalign/pipelines/asr/num2chinese.py
+batchalign/pipelines/asr/oai_whisper.py
 batchalign/pipelines/asr/rev.py
 batchalign/pipelines/asr/utils.py
 batchalign/pipelines/asr/whisper.py

{batchalign-0.7.18.post12 → batchalign-0.7.19}/batchalign.egg-info/requires.txt

@@ -24,6 +24,7 @@ tiktoken
 blobfile
 sentencepiece
 googletrans
+openai-whisper>=20240930
 
 [dev]
 pytest

{batchalign-0.7.18.post12 → batchalign-0.7.19}/setup.py

@@ -55,7 +55,8 @@ setup(
         "tiktoken",
         "blobfile",
         "sentencepiece",
-        "googletrans"
+        "googletrans",
+        "openai-whisper>=20240930"
     ],
     extras_require={
         'dev': [

batchalign-0.7.18.post12/batchalign/version

@@ -1,3 +0,0 @@
-0.7.18-post.12
-May 1st, 2025
-align