sonusai 0.17.2__tar.gz → 0.17.3__tar.gz

{sonusai-0.17.2 → sonusai-0.17.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.17.2
+Version: 0.17.3
 Summary: Framework for building deep neural network models for sound, speech, and voice AI
 Home-page: https://aaware.com
 License: GPL-3.0-only
@@ -21,6 +21,7 @@ Requires-Dist: h5py (>=3.11.0,<4.0.0)
 Requires-Dist: jiwer (>=3.0.3,<4.0.0)
 Requires-Dist: librosa (>=0.10.1,<0.11.0)
 Requires-Dist: matplotlib (>=3.8.0,<4.0.0)
+Requires-Dist: numpy (>=1.26.4,<2.0.0)
 Requires-Dist: onnx (>=1.14.1,<2.0.0)
 Requires-Dist: onnxruntime (>=1.16.1,<2.0.0)
 Requires-Dist: paho-mqtt (>=2.0.0,<3.0.0)

{sonusai-0.17.2 → sonusai-0.17.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sonusai"
-version = "0.17.2"
+version = "0.17.3"
 description = "Framework for building deep neural network models for sound, speech, and voice AI"
 authors = ["Chris Eddington <chris@aaware.com>", "Jason Calderwood <jason@aaware.com>"]
 maintainers = ["Chris Eddington <chris@aaware.com>", "Jason Calderwood <jason@aaware.com>"]
@@ -21,6 +21,7 @@ h5py = "^3.11.0"
 jiwer = "^3.0.3"
 librosa = "^0.10.1"
 matplotlib = "^3.8.0"
+numpy = "^1.26.4"
 onnx = "^1.14.1"
 onnxruntime = "^1.16.1"
 paho-mqtt = "^2.0.0"
@@ -47,6 +48,7 @@ mypy = "^1.6.0"
 mypy-extensions = "^1.0.0"
 pytest = "^8.1.1"
 sonusai-asr-cloud = "^0.1.0"
+sonusai-asr-sensory = "^0.1.0"
 sonusai-torchl = "^0.1.0"
 types-pyyaml = "^6.0.12.12"
 types-requests = "^2.31.0.8"

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/audiofe.py

@@ -86,7 +86,7 @@ def main() -> None:
     from sonusai.utils import create_timestamp
     from sonusai.utils import get_input_devices
     from sonusai.utils import load_ort_session
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     ts = create_timestamp()
     capture_name = f'audiofe_capture_{ts}'
@@ -121,7 +121,7 @@ def main() -> None:
             logger.exception(e)
             return
         # Only write if capture from device, not for file input
-        write_wav(capture_wav, capture_audio, SAMPLE_RATE)
+        write_audio(capture_wav, capture_audio, SAMPLE_RATE)
         logger.info('')
         logger.info(f'Wrote capture audio with shape {capture_audio.shape} to {capture_wav}')
 
@@ -175,7 +175,7 @@ def main() -> None:
             logger.info(f'Wrote predict with shape {predict.shape} to {h5_name}')
 
         predict_audio = get_audio_from_feature(feature=predict, feature_mode=feature_mode)
-        write_wav(predict_wav, predict_audio, SAMPLE_RATE)
+        write_audio(predict_wav, predict_audio, SAMPLE_RATE)
         logger.info(f'Wrote predict audio with shape {predict_audio.shape} to {predict_wav}')
         if debug:
             with h5py.File(h5_name, 'a') as f:

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/calc_metric_spenh.py

@@ -718,7 +718,7 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
     from sonusai.utils import reshape_outputs
     from sonusai.utils import stack_complex
     from sonusai.utils import unstack_complex
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     mixdb = MP_GLOBAL.mixdb
     predict_location = MP_GLOBAL.predict_location
@@ -968,12 +968,12 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
 
     # 7) write wav files
     if enable_wav:
-        write_wav(name=base_name + '_mixture.wav', audio=float_to_int16(mixture))
-        write_wav(name=base_name + '_target.wav', audio=float_to_int16(target))
-        # write_wav(name=base_name + '_target_fi.wav', audio=float_to_int16(target_fi))
-        write_wav(name=base_name + '_noise.wav', audio=float_to_int16(noise))
-        write_wav(name=base_name + '_target_est.wav', audio=float_to_int16(target_est_wav))
-        write_wav(name=base_name + '_noise_est.wav', audio=float_to_int16(noise_est_wav))
+        write_audio(name=base_name + '_mixture.wav', audio=float_to_int16(mixture))
+        write_audio(name=base_name + '_target.wav', audio=float_to_int16(target))
+        # write_audio(name=base_name + '_target_fi.wav', audio=float_to_int16(target_fi))
+        write_audio(name=base_name + '_noise.wav', audio=float_to_int16(noise))
+        write_audio(name=base_name + '_target_est.wav', audio=float_to_int16(target_est_wav))
+        write_audio(name=base_name + '_noise_est.wav', audio=float_to_int16(noise_est_wav))
 
         # debug code to test for perfect reconstruction of the extraction method
         # note both 75% olsa-hanns and 50% olsa-hann modes checked to have perfect reconstruction

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/mixture/mixdb.py

@@ -10,7 +10,6 @@ from typing import Optional
 
 from praatio import textgrid
 from praatio.utilities.constants import Interval
-
 from sonusai.mixture.datatypes import AudioF
 from sonusai.mixture.datatypes import AudioT
 from sonusai.mixture.datatypes import AudiosF

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/mixture/sox_augmentation.py

@@ -84,6 +84,7 @@ def apply_impulse_response(audio: AudioT, ir: ImpulseResponseData) -> AudioT:
     :return: Augmented audio
     """
     import math
+    from pathlib import Path
     import tempfile
 
     import numpy as np
@@ -124,7 +125,9 @@ def apply_impulse_response(audio: AudioT, ir: ImpulseResponseData) -> AudioT:
     except Exception as e:
         raise SonusAIError(f'Error applying IR: {e}')
 
+    path = Path(temp.name)
     temp.close()
+    path.unlink()
 
     # Reset level to previous max value
     tfm = Transformer()

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/mkwav.py

@@ -72,7 +72,7 @@ def _process_mixture(mixid: int) -> None:
 
     from sonusai.mixture import mixture_metadata
     from sonusai.utils import float_to_int16
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     mixture_filename = join(MP_GLOBAL.mixdb.location, MP_GLOBAL.mixdb.mixtures[mixid].name)
     mixture_basename = splitext(mixture_filename)[0]
@@ -100,11 +100,11 @@ def _process_mixture(mixid: int) -> None:
             if MP_GLOBAL.write_noise:
                 noise = np.array(f['noise'])
 
-    write_wav(name=mixture_basename + '_mixture.wav', audio=float_to_int16(mixture))
+    write_audio(name=mixture_basename + '_mixture.wav', audio=float_to_int16(mixture))
     if MP_GLOBAL.write_target:
-        write_wav(name=mixture_basename + '_target.wav', audio=float_to_int16(target))
+        write_audio(name=mixture_basename + '_target.wav', audio=float_to_int16(target))
     if MP_GLOBAL.write_noise:
-        write_wav(name=mixture_basename + '_noise.wav', audio=float_to_int16(noise))
+        write_audio(name=mixture_basename + '_noise.wav', audio=float_to_int16(noise))
 
     with open(file=mixture_basename + '.txt', mode='w') as f:
         f.write(mixture_metadata(MP_GLOBAL.mixdb, MP_GLOBAL.mixdb.mixture(mixid)))

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/onnx_predict.py

@@ -100,7 +100,7 @@ def main() -> None:
     from sonusai.utils import create_ts_name
     from sonusai.utils import load_ort_session
     from sonusai.utils import reshape_inputs
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     mixdb_path = None
     mixdb = None
@@ -201,7 +201,7 @@ def main() -> None:
                 predict = np.transpose(predict, [1, 0, 2])
                 predict_audio = get_audio_from_feature(feature=predict, feature_mode=feature_mode)
                 owav_name = splitext(output_fname)[0] + '_predict.wav'
-                write_wav(owav_name, predict_audio)
+                write_audio(owav_name, predict_audio)
 
 
 if __name__ == '__main__':

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/post_spenh_targetf.py

@@ -137,7 +137,7 @@ def _process(file: str) -> None:
     from sonusai.mixture import get_audio_from_transform
     from sonusai.utils import float_to_int16
     from sonusai.utils import unstack_complex
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     try:
         with h5py.File(file, 'r') as f:
@@ -153,7 +153,7 @@ def _process(file: str) -> None:
                                                                          bin_end=MP_GLOBAL.bin_end,
                                                                          ttype=MP_GLOBAL.ttype,
                                                                          gain=np.float32(1)))
-    write_wav(name=output_name, audio=float_to_int16(audio))
+    write_audio(name=output_name, audio=float_to_int16(audio))
 
 
 if __name__ == '__main__':

sonusai-0.17.2/sonusai/speech/voxceleb2.py → sonusai-0.17.3/sonusai/speech/voxceleb.py

@@ -16,14 +16,30 @@ def load_speakers(input_dir: Path) -> dict:
     import csv
 
     speakers = {}
+
+    # VoxCeleb1
+    first = True
+    with open(input_dir / 'vox1_meta.csv', newline='') as file:
+        data = csv.reader(file, delimiter='\t')
+        for row in data:
+            if first:
+                first = False
+            else:
+                speakers[row[0].strip()] = {'gender':   row[2].strip(),
+                                            'dialect':  row[3].strip(),
+                                            'category': row[4].strip()}
+
+    # VoxCeleb2
     first = True
-    with open(input_dir / 'vox2_meta_cleansed.csv', newline='') as file:
-        data = csv.reader(file)
+    with open(input_dir / 'vox2_meta.csv', newline='') as file:
+        data = csv.reader(file, delimiter='\t')
         for row in data:
             if first:
                 first = False
             else:
-                speakers[row[0].strip()] = {'gender': row[2].strip(), 'category': row[3].strip()}
+                speakers[row[1].strip()] = {'gender':   row[3].strip(),
+                                            'category': row[4].strip()}
+
     return speakers
 
 

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/utils/__init__.py

@@ -49,5 +49,5 @@ from .stacked_complex import stacked_complex_imag
 from .stacked_complex import stacked_complex_real
 from .stacked_complex import unstack_complex
 from .stratified_shuffle_split import stratified_shuffle_split_mixid
-from .wave import write_wav
+from .write_audio import write_audio
 from .yes_or_no import yes_or_no

{sonusai-0.17.2 → sonusai-0.17.3}/sonusai/utils/asr_functions/aaware_whisper.py

@@ -13,7 +13,7 @@ def aaware_whisper(data: ASRData) -> ASRResult:
     from sonusai import SonusAIError
     from sonusai.utils import ASRResult
     from sonusai.utils import float_to_int16
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     url = getenv('AAWARE_WHISPER_URL')
     if url is None:
@@ -22,7 +22,7 @@ def aaware_whisper(data: ASRData) -> ASRResult:
 
     with tempfile.TemporaryDirectory() as tmp:
         file = join(tmp, 'asr.wav')
-        write_wav(name=file, audio=float_to_int16(data.audio))
+        write_audio(name=file, audio=float_to_int16(data.audio))
 
         files = {'audio_file': (file, open(file, 'rb'), 'audio/wav')}
 

sonusai-0.17.2/sonusai/utils/wave.py → sonusai-0.17.3/sonusai/utils/write_audio.py

@@ -2,8 +2,8 @@ from sonusai.mixture.constants import SAMPLE_RATE
 from sonusai.mixture.datatypes import AudioT
 
 
-def write_wav(name: str, audio: AudioT, sample_rate: int = SAMPLE_RATE) -> None:
-    """ Write a simple, uncompressed WAV file.
+def write_audio(name: str, audio: AudioT, sample_rate: int = SAMPLE_RATE) -> None:
+    """ Write an audio file.
 
     To write multiple channels, use a 2D array of shape [channels, samples].
     The bits per sample and PCM/float are determined by the data type.