lattifai 0.2.2__py3-none-any.whl → 0.2.4__py3-none-any.whl

lattifai/workers/lattice1_alpha.py

@@ -9,6 +9,7 @@ import resampy
 import soundfile as sf
 import torch
 from lhotse import FbankConfig
+from lhotse.audio import read_audio
 from lhotse.features.kaldi.layers import Wav2LogFilterBank
 from lhotse.utils import Pathlike
 
@@ -76,13 +77,55 @@ class Lattice1AlphaWorker:
 
     def load_audio(self, audio: Union[Pathlike, BinaryIO]) -> Tuple[torch.Tensor, int]:
         # load audio
-        waveform, sample_rate = sf.read(audio, always_2d=True, dtype='float32')
-        if waveform.shape[1] > 1:  # TODO: support choose channel
-            waveform = np.mean(waveform, axis=1, keepdims=True)
+        try:
+            waveform, sample_rate = read_audio(audio)  # numpy array
+            if len(waveform.shape) == 1:
+                waveform = waveform.reshape([1, -1])  # (1, L)
+            else:  # make sure channel first
+                if waveform.shape[0] > waveform.shape[1]:
+                    waveform = waveform.transpose(0, 1)
+                # average multiple channels
+                waveform = np.mean(waveform, axis=0, keepdims=True)  # (1, L)
+        except Exception:
+            # Fallback to PyAV for formats not supported by soundfile
+            import av
+
+            container = av.open(audio)
+            audio_stream = next((s for s in container.streams if s.type == 'audio'), None)
+
+            if audio_stream is None:
+                raise ValueError(f'No audio stream found in {audio}')
+
+            # Resample to target sample rate during decoding
+            audio_stream.codec_context.format = av.AudioFormat('flt')  # 32-bit float
+
+            frames = []
+            for frame in container.decode(audio_stream):
+                # Convert frame to numpy array
+                array = frame.to_ndarray()
+                # Ensure shape is (channels, samples)
+                if array.ndim == 1:
+                    array = array.reshape(1, -1)
+                elif array.ndim == 2 and array.shape[0] > array.shape[1]:
+                    array = array.T
+                frames.append(array)
+
+            container.close()
+
+            if not frames:
+                raise ValueError(f'No audio data found in {audio}')
+
+            # Concatenate all frames
+            waveform = np.concatenate(frames, axis=1)
+            # Average multiple channels to mono
+            if waveform.shape[0] > 1:
+                waveform = np.mean(waveform, axis=0, keepdims=True)
+
+            sample_rate = audio_stream.codec_context.sample_rate
+
         if sample_rate != self.config['sample_rate']:
-            waveform = resampy.resample(waveform, sample_rate, self.config['sample_rate'], axis=0)
-        waveform = torch.from_numpy(waveform.T).to(self.device)  # (1, L)
-        return waveform
+            waveform = resampy.resample(waveform, sample_rate, self.config['sample_rate'], axis=1)
+        return torch.from_numpy(waveform).to(self.device)  # (1, L)
 
     def alignment(
         self, audio: Union[Union[Pathlike, BinaryIO], torch.tensor], lattice_graph: Tuple[str, int, float]

{lattifai-0.2.2.dist-info → lattifai-0.2.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lattifai
-Version: 0.2.2
+Version: 0.2.4
 Summary: Lattifai Python SDK: Seamless Integration with Lattifai's Speech and Video AI Services
 Author-email: Lattifai Technologies <tech@lattifai.com>
 Maintainer-email: Lattice <tech@lattifai.com>
@@ -61,6 +61,7 @@ Requires-Dist: onnxruntime
 Requires-Dist: resampy
 Requires-Dist: g2p-phonemizer==0.1.1
 Requires-Dist: wtpsplit>=2.1.6
+Requires-Dist: av
 Provides-Extra: numpy
 Requires-Dist: numpy; extra == "numpy"
 Provides-Extra: test

{lattifai-0.2.2.dist-info → lattifai-0.2.4.dist-info}/RECORD

@@ -13,10 +13,10 @@ lattifai/tokenizer/__init__.py,sha256=aqv44PDtq6g3oFFKW_l4HSR5ywT5W8eP1dHHywIvBf
 lattifai/tokenizer/phonemizer.py,sha256=SfRi1KIMpmaao6OVmR1h_I_3QU-vrE6D5bh72Afg5XM,1759
 lattifai/tokenizer/tokenizer.py,sha256=Yuo0pLPQnF2uX0Fm5g8i5vtcADn7GeLpSqdGpMJgTww,11492
 lattifai/workers/__init__.py,sha256=s6YfkIq4FDIAzY9sPjRpXnJfszj2repqnMTqydRM5Zw,83
-lattifai/workers/lattice1_alpha.py,sha256=1VFo59EcygEctTHOhkcII8v3_mrj8JEJ8Fcaqk_7LVo,5762
-lattifai-0.2.2.dist-info/licenses/LICENSE,sha256=LNuoH5jpXXNKgjQ3XLwztFq8D3O7kZI-LSg81o4ym2M,1065
-lattifai-0.2.2.dist-info/METADATA,sha256=4vmPOYKsIlvADiw0zUDQ2dbDpe-vOV-o5A0Hs1p7xfg,10971
-lattifai-0.2.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lattifai-0.2.2.dist-info/entry_points.txt,sha256=CwTI2NbJvF9msIHboAfTA99cmDr_HOWoODjS8R64JOw,131
-lattifai-0.2.2.dist-info/top_level.txt,sha256=-OVWZ68YYFcTN13ARkLasp2OUappe9wEVq-CKes7jM4,17
-lattifai-0.2.2.dist-info/RECORD,,
+lattifai/workers/lattice1_alpha.py,sha256=fnimZqhPQ1fBCUjcDVblnFGTWP0vbNLRM7E7lOdHJu8,7428
+lattifai-0.2.4.dist-info/licenses/LICENSE,sha256=LNuoH5jpXXNKgjQ3XLwztFq8D3O7kZI-LSg81o4ym2M,1065
+lattifai-0.2.4.dist-info/METADATA,sha256=7SNTA_Egpv3F5rd0F20-4MigC7muz5x6kyr_xxj4CIk,10989
+lattifai-0.2.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lattifai-0.2.4.dist-info/entry_points.txt,sha256=CwTI2NbJvF9msIHboAfTA99cmDr_HOWoODjS8R64JOw,131
+lattifai-0.2.4.dist-info/top_level.txt,sha256=-OVWZ68YYFcTN13ARkLasp2OUappe9wEVq-CKes7jM4,17
+lattifai-0.2.4.dist-info/RECORD,,