BatchalignHK 0.7.22.post10__tar.gz → 0.7.22.post12__tar.gz

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/BatchalignHK.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: BatchalignHK
-Version: 0.7.22.post10
+Version: 0.7.22.post12
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -48,6 +48,7 @@ Requires-Dist: cos-python-sdk-v5
 Requires-Dist: openai-whisper
 Requires-Dist: llvmlite>=0.44.0
 Requires-Dist: praat-parselmouth==0.4.6
+Requires-Dist: opensmile>=2.3.0
 Requires-Dist: pyannote.audio
 Requires-Dist: onnxruntime
 Provides-Extra: dev

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/BatchalignHK.egg-info/SOURCES.txt

@@ -123,6 +123,8 @@ batchalign/pipelines/morphosyntax/fr/apm.py
 batchalign/pipelines/morphosyntax/fr/apmn.py
 batchalign/pipelines/morphosyntax/fr/case.py
 batchalign/pipelines/morphosyntax/ja/verbforms.py
+batchalign/pipelines/opensmile/__init__.py
+batchalign/pipelines/opensmile/engine.py
 batchalign/pipelines/speaker/__init__.py
 batchalign/pipelines/speaker/nemo_speaker.py
 batchalign/pipelines/translate/__init__.py

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/BatchalignHK.egg-info/requires.txt

@@ -38,6 +38,7 @@ cos-python-sdk-v5
 openai-whisper
 llvmlite>=0.44.0
 praat-parselmouth==0.4.6
+opensmile>=2.3.0
 pyannote.audio
 onnxruntime
 

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: BatchalignHK
-Version: 0.7.22.post10
+Version: 0.7.22.post12
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu
@@ -48,6 +48,7 @@ Requires-Dist: cos-python-sdk-v5
 Requires-Dist: openai-whisper
 Requires-Dist: llvmlite>=0.44.0
 Requires-Dist: praat-parselmouth==0.4.6
+Requires-Dist: opensmile>=2.3.0
 Requires-Dist: pyannote.audio
 Requires-Dist: onnxruntime
 Provides-Extra: dev

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/batchalign/cli/cli.py

@@ -425,7 +425,41 @@ def avqi(ctx, cs_file, sv_file, lang, **kwargs):
             import traceback
             C.print(traceback.format_exc())
 
+#################### OPENSMILE ################################
 
+@batchalign.command()
+@click.argument("input_dir", type=click.Path(exists=True, file_okay=False))
+@click.argument("output_dir", type=click.Path(exists=True, file_okay=False))
+@click.option("--feature-set", 
+              type=click.Choice(['eGeMAPSv02', 'eGeMAPSv01b', 'GeMAPSv01b', 'ComParE_2016']),
+              default='eGeMAPSv02',
+              help="Feature set to extract")
+@click.option("--lang",
+              help="sample language in three-letter ISO 3166-1 alpha-3 code",
+              show_default=True, default="eng", type=str)
+@click.pass_context
+def opensmile(ctx, input_dir, output_dir, feature_set, lang, **kwargs):
+    """Extract openSMILE audio features from speech samples."""
+
+    def loader(file):
+        doc = Document.new(media_path=file, lang=lang)
+        return doc, {"feature_set": feature_set}
+
+    def writer(results, output):
+        if results.get('success', False):
+            output_csv = Path(output).with_suffix('.opensmile.csv')
+            features_df = results.get('features_df')
+            if features_df is not None:
+                features_df.to_csv(output_csv, header=['value'], index_label='feature')
+        else:
+            error_file = Path(output).with_suffix('.error.txt')
+            with open(error_file, 'w') as f:
+                f.write(f"OpenSMILE extraction failed: {results.get('error', 'Unknown error')}\n")
+
+    _dispatch("opensmile", lang, 1, ["mp3", "mp4", "wav"], ctx,
+              input_dir, output_dir,
+              loader, writer, C, **kwargs)
+    
 #################### SETUP ################################
 
 @batchalign.command()
@@ -447,4 +481,3 @@ def version(ctx, **kwargs):
            f"[italic]{RELEASE_NOTES.strip()}[/italic]"+"\n" +
            "\nDeveloped by Brian MacWhinney and Houjun Liu")
     C.print("\n\n"+ptr+"\n\n")
-

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/batchalign/cli/dispatch.py

@@ -50,6 +50,7 @@ Cmd2Task = {
     "utseg": "utterance",
     "coref": "coref",
     "translate": "translate",
+    "opensmile": "opensmile",
 }
 
 # this is the main runner used by all functions

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/batchalign/pipelines/__init__.py

@@ -17,3 +17,4 @@ from .translate import SeamlessTranslationModel, GoogleTranslateEngine
 from .avqi import AVQIEngine
 
 from .diarization import PyannoteEngine
+from .opensmile import OpenSMILEEngine

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/batchalign/pipelines/asr/utils.py

@@ -60,10 +60,11 @@ def retokenize(intermediate_output):
             word = word.replace("。", ".")
             word = word.replace("¿", " ").replace("¡", " ")
             tmp.append((word, bullet))
-            if len(word) > 0 and (word in ENDING_PUNCT or word[-1] in ENDING_PUNCT):
-                if word in ENDING_PUNCT:
+            if len(word) > 0 and (word in ENDING_PUNCT+["؟", "۔", "،", "؛"]
+                                  or word[-1] in ENDING_PUNCT+["؟", "۔", "،", "؛"]):
+                if word in ENDING_PUNCT+["؟", "۔", "،", "؛"]:
                     final_outputs.append((speaker, tmp))
-                elif word[-1] in ENDING_PUNCT:
+                elif word[-1] in ENDING_PUNCT+["؟", "۔", "،", "؛"]:
                     # we want to seperate the ending punct out
                     final, time = tmp.pop(-1)
                     tmp.append((final[:-1], time))
@@ -102,7 +103,7 @@ def retokenize_with_engine(intermediate_output, engine):
         # because we are using an utterance engine, we need
         # to get rid of all the preexisting punctuation
         for i in utterance:
-            for j in MOR_PUNCT+ENDING_PUNCT:
+            for j in MOR_PUNCT+ENDING_PUNCT+["؟", "۔", "،", "؛"]:
                 i[0] = i[0].strip(j).lower()
 
         # remove everything that's now blank
@@ -118,7 +119,7 @@ def retokenize_with_engine(intermediate_output, engine):
         # align the utterance against original splits and generate final outputs
         for i in split:
             # Check if the split has ending punctuation
-            if i[-1] in ENDING_PUNCT:
+            if i[-1] in ENDING_PUNCT+["؟", "۔", "،", "؛"]:
                 new_ut, delim = (i[:-1].split(" "), i[-1])
             else:
                 new_ut, delim = (i.split(" "), ".")
@@ -273,16 +274,8 @@ def process_generation(output, lang="eng", utterance_engine=None):
                 seen_word = False
             if word.strip() == "":
                 continue
-            if word not in ENDING_PUNCT+MOR_PUNCT:
+            if word not in ENDING_PUNCT+MOR_PUNCT+["؟", "۔", "،", "؛"]:
                 word_replaced = word
-                if word_replaced.strip() == "؟":
-                    word_replaced = "?"
-                elif word_replaced.strip() == "۔":
-                    word_replaced = "."
-                elif word_replaced.strip() == "،":
-                    word_replaced = ","
-                elif word_replaced.strip() == "؛":
-                    word_replaced = ";"
                     
                 if start == None or end == None:
                     words.append(Form(text=word_replaced, time=None))
@@ -290,7 +283,15 @@ def process_generation(output, lang="eng", utterance_engine=None):
                     seen_word = True
                     words.append(Form(text=word_replaced, time=(int(start), int(end))))
             else:
-                    words.append(Form(text=word, time=None))
+                if word.strip() == "؟":
+                    word = "?"
+                elif word.strip() == "۔":
+                    word = "."
+                elif word.strip() == "،":
+                    word = ","
+                elif word.strip() == "؛":
+                    word = ";"
+                words.append(Form(text=word, time=None))
 
         final_utterances.append(Utterance(
             tier=participant,

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/batchalign/pipelines/dispatch.py

@@ -33,6 +33,7 @@ DEFAULT_PACKAGES = {
     "utterance": "stanza_utt",
     "coref": "stanza_coref",
     "translate": "gtrans",
+    "opensmile": "opensmile_egemaps",
 }
 
 LANGUAGE_OVERRIDE_PACKAGES = {
@@ -152,8 +153,19 @@ def dispatch_pipeline(pkg_str, lang, num_speakers=None, **arg_overrides):
             engines.append(FunAudioUTREngine(lang=lang))
         elif engine == "pyannote":
             engines.append(PyannoteEngine())
+        elif engine == "opensmile_egemaps":
+            from batchalign.pipelines.opensmile import OpenSMILEEngine
+            engines.append(OpenSMILEEngine(feature_set='eGeMAPSv02'))
+        elif engine == "opensmile_gemaps":
+            from batchalign.pipelines.opensmile import OpenSMILEEngine
+            engines.append(OpenSMILEEngine(feature_set='GeMAPSv01b'))
+        elif engine == "opensmile_compare":
+            from batchalign.pipelines.opensmile import OpenSMILEEngine
+            engines.append(OpenSMILEEngine(feature_set='ComParE_2016'))
+        elif engine == "opensmile_eGeMAPSv01b":
+            from batchalign.pipelines.opensmile import OpenSMILEEngine
+            engines.append(OpenSMILEEngine(feature_set='eGeMAPSv01b'))
 
 
     L.debug(f"Done initalizing packages.")
     return BatchalignPipeline(*engines)
-    

batchalignhk-0.7.22.post12/batchalign/pipelines/opensmile/__init__.py

@@ -0,0 +1,7 @@
+"""
+OpenSMILE Pipeline Module
+"""
+
+from .engine import OpenSMILEEngine
+
+__all__ = ['OpenSMILEEngine']

batchalignhk-0.7.22.post12/batchalign/pipelines/opensmile/engine.py

@@ -0,0 +1,191 @@
+"""
+OpenSMILE Engine for Batchalign2 - M1 Mac Compatible Version
+Audio feature extraction using the openSMILE toolkit
+"""
+
+import opensmile
+import pandas as pd
+from pathlib import Path
+import logging
+from typing import Dict, Optional
+import platform
+
+from batchalign.pipelines.base import BatchalignEngine
+from batchalign.document import Task, TaskType, Document
+
+L = logging.getLogger('batchalign')
+
+class OpenSMILEEngine(BatchalignEngine):
+    """Engine for extracting openSMILE audio features."""
+
+    def __init__(self, feature_set: str = 'eGeMAPSv02', 
+                 feature_level: str = 'functionals'):
+        super().__init__()
+        self._tasks = [Task.FEATURE_EXTRACT]
+
+        self.feature_set = feature_set
+        self.feature_level = feature_level
+
+        self.is_m1_mac = (platform.system() == 'Darwin' and 
+                         platform.processor() == 'arm')
+
+        try:
+            if self.is_m1_mac:
+                L.info("M1 Mac detected - using default openSMILE configuration")
+                self.smile = opensmile.Smile()
+                self._requested_feature_set = feature_set
+            else:
+                self.smile = opensmile.Smile(
+                    feature_set=feature_set,
+                    feature_level=feature_level,
+                )
+            L.debug(f"OpenSMILE initialized (M1 compatibility mode: {self.is_m1_mac})")
+        except Exception as e:
+            L.error(f"Failed to initialize openSMILE: {e}")
+            raise
+
+    @property
+    def tasks(self):
+        return self._tasks
+
+    def analyze(self, doc: Document, feature_set: str = None, **kwargs) -> Dict:
+        """
+        Extract openSMILE features from Document.
+        
+        Args:
+            doc: Document with media attached
+            feature_set: Feature set to use (ignored on M1 Mac)
+            **kwargs: Additional arguments
+            
+        Returns:
+            Dictionary with extraction results and metadata
+        """
+
+        if not doc.media or not doc.media.url:
+            return {
+                'error': 'Document has no media attached',
+                'success': False
+            }
+
+        actual_audio_path = doc.media.url
+
+        if feature_set and feature_set != self.feature_set:
+            if self.is_m1_mac:
+                L.warning(f"Feature set switching not supported on M1 Mac - using default features instead of {feature_set}")
+            else:
+                L.info(f"Switching feature set from {self.feature_set} to {feature_set}")
+                try:
+                    self.feature_set = feature_set
+                    self.smile = opensmile.Smile(
+                        feature_set=feature_set,
+                        feature_level=self.feature_level,
+                    )
+                except Exception as e:
+                    L.error(f"Failed to switch to feature set {feature_set}: {e}")
+                    return {
+                        'feature_set': self.feature_set,
+                        'num_features': 0,
+                        'error': f"Feature set switch failed: {str(e)}",
+                        'success': False
+                    }
+
+        try:
+            L.info(f"Extracting features from: {Path(actual_audio_path).name}")
+            if self.is_m1_mac:
+                L.info("Using M1-compatible default feature set (eGeMAPSv02 equivalent)")
+            else:
+                L.info(f"Using {self.feature_set} feature set")
+
+            features_df = self.smile.process_file(actual_audio_path)
+
+            if features_df is None or features_df.empty:
+                raise ValueError("Feature extraction returned empty results")
+
+            results_df = features_df.T
+
+            num_features = len(features_df.columns)
+            duration_segments = len(features_df)
+
+            first_row_features = {}
+            if duration_segments > 0:
+                first_row_features = features_df.iloc[0].to_dict()
+
+            actual_feature_set = self.feature_set
+            if self.is_m1_mac:
+                actual_feature_set = "M1-default (eGeMAPSv02-like)"
+
+            results = {
+                'feature_set': actual_feature_set,
+                'feature_level': self.feature_level,
+                'num_features': num_features,
+                'duration_segments': duration_segments,
+                'audio_file': str(actual_audio_path),
+                'features_sample': first_row_features,
+                'success': True,
+                'm1_compatibility_mode': self.is_m1_mac,
+                'features_df': results_df,
+            }
+
+            if self.is_m1_mac and hasattr(self, '_requested_feature_set'):
+                results['requested_feature_set'] = self._requested_feature_set
+                results['warning'] = f"M1 Mac compatibility: used default features instead of {self._requested_feature_set}"
+
+            L.info(f"Successfully extracted {num_features} features from {duration_segments} segments")
+            return results
+
+        except Exception as e:
+            L.error(f"Error extracting openSMILE features from {actual_audio_path}: {e}")
+            return {
+                'feature_set': self.feature_set,
+                'feature_level': self.feature_level,
+                'num_features': 0,
+                'duration_segments': 0,
+                'audio_file': str(actual_audio_path),
+                'error': str(e),
+                'success': False,
+                'm1_compatibility_mode': self.is_m1_mac
+            }
+
+    def get_available_feature_sets(self) -> list:
+        """Return list of available feature sets (limited on M1 Mac)."""
+        if self.is_m1_mac:
+            return ['M1-default (eGeMAPSv02-like)']
+        return [
+            'eGeMAPSv02',
+            'eGeMAPSv01b', 
+            'GeMAPSv01b',
+            'ComParE_2016'
+        ]
+
+    def get_feature_set_info(self, feature_set: str) -> dict:
+        """Get information about a specific feature set."""
+        if self.is_m1_mac:
+            return {
+                'description': 'M1 Mac compatible default feature set (similar to eGeMAPSv02)',
+                'num_features': 'Variable',
+                'recommended_for': 'General audio analysis on Apple Silicon'
+            }
+
+        info = {
+            'eGeMAPSv02': {
+                'description': 'Extended Geneva Minimalistic Acoustic Parameter Set v02',
+                'num_features': 88,
+                'recommended_for': 'General emotion and paralinguistic analysis'
+            },
+            'eGeMAPSv01b': {
+                'description': 'Extended Geneva Minimalistic Acoustic Parameter Set v01b', 
+                'num_features': 88,
+                'recommended_for': 'Emotion recognition, clinical assessment'
+            },
+            'GeMAPSv01b': {
+                'description': 'Geneva Minimalistic Acoustic Parameter Set v01b',
+                'num_features': 62,
+                'recommended_for': 'Basic paralinguistic analysis'
+            },
+            'ComParE_2016': {
+                'description': 'Computational Paralinguistics Challenge 2016 feature set',
+                'num_features': 6373,
+                'recommended_for': 'Comprehensive analysis (large feature space)'
+            }
+        }
+        return info.get(feature_set, {'description': 'Unknown feature set', 'num_features': 'Unknown'})

batchalignhk-0.7.22.post12/batchalign/version

@@ -0,0 +1,3 @@
+0.7.22-post.12
+October 10st, 2025
+OpenSMILE Analysis

{batchalignhk-0.7.22.post10 → batchalignhk-0.7.22.post12}/setup.py

@@ -69,7 +69,8 @@ setup(
         "cos-python-sdk-v5",
         "openai-whisper",
         "llvmlite>=0.44.0",
-        "praat-parselmouth==0.4.6",  # Added parselmouth for AVQI
+        "praat-parselmouth==0.4.6",  
+        "opensmile>=2.3.0",
         "pyannote.audio",
         "onnxruntime"
     ],

batchalignhk-0.7.22.post10/batchalign/version

@@ -1,3 +0,0 @@
-0.7.22-post.10
-October 1st, 2025
-Count initalisms only when its caps