Merge pull request #64 from alibaba-damo-academy/dev_lhn

Dev lhn

funasr/datasets/iterable_dataset.py

@@ -20,7 +20,7 @@ import os.path
 from funasr.datasets.dataset import ESPnetDataset
 
 
-SUPPORT_AUDIO_TYPE_SETS = ['flac', 'mp3', 'm4a', 'ogg', 'opus', 'wav', 'wma']
+SUPPORT_AUDIO_TYPE_SETS = ['flac', 'mp3', 'ogg', 'opus', 'wav', 'pcm']
 
 def load_kaldi(input):
     retval = kaldiio.load_mat(input)
@@ -60,9 +60,14 @@ def load_bytes(input):
     array = np.frombuffer((middle_data.astype(dtype) - offset) / abs_max, dtype=np.float32)
     return array
 
+def load_pcm(input):
+    with open(input,"rb") as f:
+        bytes = f.read()
+    return load_bytes(bytes)
 
 DATA_TYPES = {
     "sound": lambda x: torchaudio.load(x)[0][0].numpy(),
+    "pcm": load_pcm,
     "kaldi_ark": load_kaldi,
     "bytes": load_bytes,
     "waveform": lambda x: x,
@@ -219,6 +224,9 @@ class IterableESPnetDataset(IterableDataset):
                 if audio_type not in SUPPORT_AUDIO_TYPE_SETS:
                     raise NotImplementedError(
                         f'Not supported audio type: {audio_type}')
+                if audio_type == "pcm":
+                    _type = "pcm"
+
             func = DATA_TYPES[_type]
             array = func(value)
             if self.fs is not None and name == "speech":
@@ -318,6 +326,8 @@ class IterableESPnetDataset(IterableDataset):
                         if audio_type not in SUPPORT_AUDIO_TYPE_SETS:
                             raise NotImplementedError(
                                 f'Not supported audio type: {audio_type}')
+                        if audio_type == "pcm":
+                            _type = "pcm"
                     func = DATA_TYPES[_type]
                     # Load entry
                     array = func(value)

funasr/utils/asr_utils.py

@@ -18,7 +18,7 @@ end_color = '\033[0m'
 
 global_asr_language = 'zh-cn'
 
-SUPPORT_AUDIO_TYPE_SETS = ['flac', 'mp3', 'm4a', 'ogg', 'opus', 'wav', 'wma']
+SUPPORT_AUDIO_TYPE_SETS = ['flac', 'mp3', 'ogg', 'opus', 'wav', 'pcm']
 
 def get_version():
     return float(pkg_resources.get_distribution('easyasr').version)
@@ -128,7 +128,12 @@ def get_sr_from_bytes(wav: bytes):
 def get_sr_from_wav(fname: str):
     fs = None
     if os.path.isfile(fname):
-        audio, fs = torchaudio.load(fname)
+        audio_type = os.path.basename(fname).split(".")[1].lower()
+        if audio_type in SUPPORT_AUDIO_TYPE_SETS:
+            if audio_type == "pcm":
+                fs = None
+            else:
+                audio, fs = torchaudio.load(fname)
         return fs
     elif os.path.isdir(fname):
         dir_files = os.listdir(fname)
@@ -347,4 +352,3 @@ def print_progress(percent):
         percent = 1
     res = int(50 * percent) * '#'
     print('\r[%-50s] %d%%' % (res, int(100 * percent)), end='')
-