ailia-speech 1.3.2.2__tar.gz → 1.4.0__tar.gz

{ailia_speech-1.3.2.2/ailia_speech.egg-info → ailia_speech-1.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.1
+Metadata-Version: 2.4
 Name: ailia_speech
-Version: 1.3.2.2
+Version: 1.4.0
 Summary: ailia AI Speech
 Home-page: https://ailia.jp/
 Author: ax Inc.
@@ -10,6 +10,15 @@ Requires-Python: >3.6
 Description-Content-Type: text/markdown
 Requires-Dist: ailia
 Requires-Dist: ailia_tokenizer
+Dynamic: author
+Dynamic: author-email
+Dynamic: description
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: license
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
 
 # ailia AI Speech Python API
 
@@ -103,13 +112,21 @@ for i in range(0, audio_waveform.shape[0], sampling_rate):
 		print(text)
 ```
 
+### Dialization mode
+
+By specifying dialization_type, speaker diarization can be performed. When speaker diarization is enabled, speaker_id becomes valid.
+
+```
+speech.initialize_model(model_path = "./models/", model_type = ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE_V3_TURBO, dialization_type = AILIA_SPEECH_DIARIZATION_TYPE_PYANNOTE_AUDIO)
+```
+
 ### Available model types
 
 It is possible to select multiple models according to accuracy and speed. LARGE_V3_TURBO is the most recommended.
 
 ```
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_TINY
-ilia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_BASE
+ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_BASE
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_SMALL
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_MEDIUM
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE

ailia_speech-1.3.2.2/PKG-INFO → ailia_speech-1.4.0/README.md

@@ -1,16 +1,3 @@
-Metadata-Version: 2.1
-Name: ailia_speech
-Version: 1.3.2.2
-Summary: ailia AI Speech
-Home-page: https://ailia.jp/
-Author: ax Inc.
-Author-email: contact@axinc.jp
-License: https://ailia.ai/en/license/
-Requires-Python: >3.6
-Description-Content-Type: text/markdown
-Requires-Dist: ailia
-Requires-Dist: ailia_tokenizer
-
 # ailia AI Speech Python API
 
 !! CAUTION !!
@@ -103,13 +90,21 @@ for i in range(0, audio_waveform.shape[0], sampling_rate):
 		print(text)
 ```
 
+### Dialization mode
+
+By specifying dialization_type, speaker diarization can be performed. When speaker diarization is enabled, speaker_id becomes valid.
+
+```
+speech.initialize_model(model_path = "./models/", model_type = ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE_V3_TURBO, dialization_type = AILIA_SPEECH_DIARIZATION_TYPE_PYANNOTE_AUDIO)
+```
+
 ### Available model types
 
 It is possible to select multiple models according to accuracy and speed. LARGE_V3_TURBO is the most recommended.
 
 ```
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_TINY
-ilia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_BASE
+ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_BASE
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_SMALL
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_MEDIUM
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE

{ailia_speech-1.3.2.2 → ailia_speech-1.4.0}/ailia_speech/__init__.py

@@ -76,9 +76,13 @@ AILIA_SPEECH_FLAG_NONE = (0)
 AILIA_SPEECH_FLAG_LIVE = (1)
 
 AILIA_SPEECH_VAD_TYPE_SILERO = (0)
+
+AILIA_SPEECH_DIARIZATION_TYPE_PYANNOTE_AUDIO = (0)
+
 AILIA_SPEECH_API_CALLBACK_VERSION = (6)
 
 AILIA_SPEECH_TEXT_VERSION = (2)
+AILIA_SPEECH_SPEAKER_ID_UNKNOWN = (0xFFFFFFFF)
 
 AILIA_SPEECH_USER_API_AILIA_AUDIO_GET_FRAME_LEN = CFUNCTYPE(POINTER(c_int), c_int, c_int, c_int, c_int)
 AILIA_SPEECH_USER_API_AILIA_AUDIO_GET_MEL_SPECTROGRAM = CFUNCTYPE((c_int), c_void_p, c_void_p, c_int, c_int, c_int, c_int, c_int, c_int, c_int, c_int, c_float, c_int, c_float, c_float, c_int, c_int, c_int)
@@ -215,6 +219,12 @@ dll.ailiaSpeechOpenVadFileA.argtypes = (c_void_p, c_char_p, c_int32)
 dll.ailiaSpeechOpenVadFileW.restype = c_int
 dll.ailiaSpeechOpenVadFileW.argtypes = (c_void_p, c_wchar_p, c_int32)
 
+dll.ailiaSpeechOpenDiarizationFileA.restype = c_int
+dll.ailiaSpeechOpenDiarizationFileA.argtypes = (c_void_p, c_char_p, c_char_p, c_int32)
+
+dll.ailiaSpeechOpenDiarizationFileW.restype = c_int
+dll.ailiaSpeechOpenDiarizationFileW.argtypes = (c_void_p, c_wchar_p, c_wchar_p, c_int32)
+
 dll.ailiaSpeechPushInputData.restype = c_int
 dll.ailiaSpeechPushInputData.argtypes = (c_void_p, numpy.ctypeslib.ndpointer(
                 dtype=numpy.float32, flags='CONTIGUOUS'
@@ -243,7 +253,7 @@ class AILIASpeechText(ctypes.Structure):
         ("text", ctypes.c_char_p),
         ("time_stamp_begin", ctypes.c_float),
         ("time_stamp_end", ctypes.c_float),
-        ("person_id", ctypes.c_uint),
+        ("speaker_id", ctypes.c_uint),
         ("language", ctypes.c_char_p),
         ("confidence", ctypes.c_float)]
 
@@ -399,7 +409,7 @@ class Whisper(AiliaSpeechModel):
             intermediate_callback_cnt = intermediate_callback_cnt + 1
 
 
-    def initialize_model(self, model_path = "./", model_type = AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_TINY):
+    def initialize_model(self, model_path = "./", model_type = AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_TINY, vad_type = AILIA_SPEECH_VAD_TYPE_SILERO, dialization_type = None):
         if "time_license" in ailia.get_version():
             ailia.check_and_download_license()
         if model_type == AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_TINY:
@@ -433,16 +443,20 @@ class Whisper(AiliaSpeechModel):
             encoder_pb_path = "encoder_large_v3_weights.pb"
             decoder_pb_path = "decoder_large_v3_fix_kv_cache_weights.pb"
         elif model_type == AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE_V3_TURBO:
-            encoder_path = "encoder_turbo.onnx"
-            decoder_path = "decoder_turbo_fix_kv_cache.onnx"
-            encoder_pb_path = "encoder_turbo_weights.pb"
+            encoder_path = "encoder_turbo.opt.onnx"
+            decoder_path = "decoder_turbo_fix_kv_cache.opt.onnx"
+            encoder_pb_path = "encoder_turbo_weights.opt.pb"
             decoder_pb_path = None
             model_type = AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE_V3
-        self._download_model(model_path, encoder_path, decoder_path, encoder_pb_path, decoder_pb_path)
+        self._download_model(model_path, encoder_path, decoder_path, encoder_pb_path, decoder_pb_path, vad_type, dialization_type)
         self._open_model(model_path + encoder_path, model_path + decoder_path, model_type)
-        self._open_vad(model_path + "silero_vad.onnx", AILIA_SPEECH_VAD_TYPE_SILERO)
-    
-    def _download_model(self, model_path, encoder_path, decoder_path, encoder_pb_path, decoder_pb_path):
+        if vad_type is not None:
+            self._open_vad(model_path + "silero_vad.onnx", vad_type)
+        if dialization_type is not None:
+            self._open_diarization(model_path + "segmentation.onnx", model_path + "speaker-embedding.onnx", dialization_type)
+
+
+    def _download_model(self, model_path, encoder_path, decoder_path, encoder_pb_path, decoder_pb_path, vad_type, dialization_type):
         REMOTE_PATH = "https://storage.googleapis.com/ailia-models/whisper/"
         os.makedirs(model_path, exist_ok = True)
         check_and_download_file(model_path + encoder_path, REMOTE_PATH)
@@ -452,8 +466,14 @@ class Whisper(AiliaSpeechModel):
         if decoder_pb_path is not None:
             check_and_download_file(model_path + decoder_pb_path, REMOTE_PATH)
 
-        REMOTE_PATH = "https://storage.googleapis.com/ailia-models/silero-vad/"
-        check_and_download_file(model_path + "silero_vad.onnx", REMOTE_PATH)
+        if vad_type is not None:
+            REMOTE_PATH = "https://storage.googleapis.com/ailia-models/silero-vad/"
+            check_and_download_file(model_path + "silero_vad.onnx", REMOTE_PATH)
+
+        if dialization_type is not None:
+            REMOTE_PATH = "https://storage.googleapis.com/ailia-models/pyannote-audio/"
+            check_and_download_file(model_path + "segmentation.onnx", REMOTE_PATH)
+            check_and_download_file(model_path + "speaker-embedding.onnx", REMOTE_PATH)
 
     def _open_model(self, encoder, decoder, model_type):
         p1 = self._string_buffer_aw(encoder)
@@ -472,6 +492,16 @@ class Whisper(AiliaSpeechModel):
         else:
             self._check(dll.ailiaSpeechOpenVadFileA(self._instance, p1, vad_type))
 
+    def _open_diarization(self, segmentation, embedding, diarization_type):
+        p1 = self._string_buffer_aw(segmentation)
+        p2 = self._string_buffer_aw(embedding)
+
+        if sys.platform == "win32":
+            self._check(dll.ailiaSpeechOpenDiarizationFileW(self._instance, p1, p2, diarization_type))
+        else:
+            self._check(dll.ailiaSpeechOpenDiarizationFileA(self._instance, p1, p2, diarization_type))
+
+
     def set_silent_threshold(self, silent_threshold, speech_sec, no_speech_sec):
         self._check(dll.ailiaSpeechSetSilentThreshold(self._instance, silent_threshold, speech_sec, no_speech_sec))
 
@@ -506,7 +536,7 @@ class Whisper(AiliaSpeechModel):
             for i in range(count.value):
                 text = AILIASpeechText()
                 self._check(dll.ailiaSpeechGetText(self._instance, ctypes.byref(text), AILIA_SPEECH_TEXT_VERSION, i))
-                yield {"text" : text.text.decode(), "time_stamp_begin" : text.time_stamp_begin, "time_stamp_end" : text.time_stamp_end, "person_id" : text.person_id, "language" : text.language.decode(), "confidence" : text.confidence}
+                yield {"text" : text.text.decode(), "time_stamp_begin" : text.time_stamp_begin, "time_stamp_end" : text.time_stamp_end, "speaker_id" : None if text.speaker_id == AILIA_SPEECH_SPEAKER_ID_UNKNOWN else text.speaker_id, "language" : text.language.decode(), "confidence" : text.confidence}
 
         self._check(dll.ailiaSpeechResetTranscribeState(self._instance))
 
@@ -542,7 +572,7 @@ class Whisper(AiliaSpeechModel):
             for i in range(count.value):
                 text = AILIASpeechText()
                 self._check(dll.ailiaSpeechGetText(self._instance, ctypes.byref(text), AILIA_SPEECH_TEXT_VERSION, i))
-                yield {"text" : text.text.decode(), "time_stamp_begin" : text.time_stamp_begin, "time_stamp_end" : text.time_stamp_end, "person_id" : text.person_id, "language" : text.language.decode(), "confidence" : text.confidence}
+                yield {"text" : text.text.decode(), "time_stamp_begin" : text.time_stamp_begin, "time_stamp_end" : text.time_stamp_end, "speaker_id" : None if text.speaker_id == AILIA_SPEECH_SPEAKER_ID_UNKNOWN else text.speaker_id, "language" : text.language.decode(), "confidence" : text.confidence}
 
         if complete:
             self._check(dll.ailiaSpeechResetTranscribeState(self._instance))

ailia_speech-1.3.2.2/README.md → ailia_speech-1.4.0/ailia_speech.egg-info/PKG-INFO

@@ -1,3 +1,25 @@
+Metadata-Version: 2.4
+Name: ailia_speech
+Version: 1.4.0
+Summary: ailia AI Speech
+Home-page: https://ailia.jp/
+Author: ax Inc.
+Author-email: contact@axinc.jp
+License: https://ailia.ai/en/license/
+Requires-Python: >3.6
+Description-Content-Type: text/markdown
+Requires-Dist: ailia
+Requires-Dist: ailia_tokenizer
+Dynamic: author
+Dynamic: author-email
+Dynamic: description
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: license
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
 # ailia AI Speech Python API
 
 !! CAUTION !!
@@ -90,13 +112,21 @@ for i in range(0, audio_waveform.shape[0], sampling_rate):
 		print(text)
 ```
 
+### Dialization mode
+
+By specifying dialization_type, speaker diarization can be performed. When speaker diarization is enabled, speaker_id becomes valid.
+
+```
+speech.initialize_model(model_path = "./models/", model_type = ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE_V3_TURBO, dialization_type = AILIA_SPEECH_DIARIZATION_TYPE_PYANNOTE_AUDIO)
+```
+
 ### Available model types
 
 It is possible to select multiple models according to accuracy and speed. LARGE_V3_TURBO is the most recommended.
 
 ```
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_TINY
-ilia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_BASE
+ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_BASE
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_SMALL
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_MEDIUM
 ailia_speech.AILIA_SPEECH_MODEL_TYPE_WHISPER_MULTILINGUAL_LARGE

{ailia_speech-1.3.2.2 → ailia_speech-1.4.0}/setup.py

@@ -54,7 +54,7 @@ if __name__ == "__main__":
     setup(
         name="ailia_speech",
         scripts=scripts,
-        version="1.3.2.2",
+        version="1.4.0",
         install_requires=[
             "ailia",
             "ailia_tokenizer",