support onnxruntime of streaming vad & bug fix

funasr/runtime/python/onnxruntime/demo_vad_offline.py

@@ -0,0 +1,11 @@
+import soundfile
+from funasr_onnx.vad_bin import Fsmn_vad
+
+
+model_dir = "/mnt/ailsa.zly/tfbase/espnet_work/FunASR_dev_zly/export/damo/speech_fsmn_vad_zh-cn-16k-common-pytorch"
+wav_path = "/mnt/ailsa.zly/tfbase/espnet_work/FunASR_dev_zly/egs_modelscope/vad/speech_fsmn_vad_zh-cn-16k-common/vad_example_16k.wav"
+model = Fsmn_vad(model_dir)
+
+#offline vad
+result = model(wav_path)
+print(result)

funasr/runtime/python/onnxruntime/demo_vad.py → funasr/runtime/python/onnxruntime/demo_vad_online.py

@@ -1,21 +1,18 @@
 import soundfile
-from funasr_onnx import Fsmn_vad
+from funasr_onnx.vad_online_bin import Fsmn_vad
 
 
-model_dir = "/Users/zhifu/Downloads/speech_fsmn_vad_zh-cn-16k-common-pytorch"
-wav_path = "/Users/zhifu/Downloads/speech_fsmn_vad_zh-cn-16k-common-pytorch/example/vad_example.wav"
+model_dir = "/mnt/ailsa.zly/tfbase/espnet_work/FunASR_dev_zly/export/damo/speech_fsmn_vad_zh-cn-16k-common-pytorch"
+wav_path = "/mnt/ailsa.zly/tfbase/espnet_work/FunASR_dev_zly/egs_modelscope/vad/speech_fsmn_vad_zh-cn-16k-common/vad_example_16k.wav"
 model = Fsmn_vad(model_dir)
 
-#offline vad
-# result = model(wav_path)
-# print(result)
 
-#online vad
+##online vad
 speech, sample_rate = soundfile.read(wav_path)
 speech_length = speech.shape[0]
-
+#
 sample_offset = 0
-step = 160 * 10
+step = 1600
 param_dict = {'in_cache': []}
 for sample_offset in range(0, speech_length, min(step, speech_length - sample_offset)):
     if sample_offset + step >= speech_length - 1:
@@ -26,5 +23,6 @@ for sample_offset in range(0, speech_length, min(step, speech_length - sample_of
     param_dict['is_final'] = is_final
     segments_result = model(audio_in=speech[sample_offset: sample_offset + step],
                             param_dict=param_dict)
-    print(segments_result)
+    if segments_result:
+        print(segments_result)
 

funasr/runtime/python/onnxruntime/funasr_onnx/utils/e2e_vad.py

@@ -439,10 +439,9 @@ class E2EVadModel():
                         - 1)) / self.vad_opts.noise_frame_num_used_for_snr
 
         return frame_state
-     
 
     def __call__(self, score: np.ndarray, waveform: np.ndarray,
-                is_final: bool = False, max_end_sil: int = 800
+                is_final: bool = False, max_end_sil: int = 800, online: bool = False
                 ):
         self.max_end_sil_frame_cnt_thresh = max_end_sil - self.vad_opts.speech_to_sil_time_thres
         self.waveform = waveform  # compute decibel for each frame
@@ -457,20 +456,29 @@ class E2EVadModel():
             segment_batch = []
             if len(self.output_data_buf) > 0:
                 for i in range(self.output_data_buf_offset, len(self.output_data_buf)):
-                    if not self.output_data_buf[i].contain_seg_start_point:
-                        continue
-                    if not self.next_seg and not self.output_data_buf[i].contain_seg_end_point:
-                        continue
-                    start_ms = self.output_data_buf[i].start_ms if self.next_seg else -1
-                    if self.output_data_buf[i].contain_seg_end_point:
+                    if online:
+                        if not self.output_data_buf[i].contain_seg_start_point:
+                            continue
+                        if not self.next_seg and not self.output_data_buf[i].contain_seg_end_point:
+                            continue
+                        start_ms = self.output_data_buf[i].start_ms if self.next_seg else -1
+                        if self.output_data_buf[i].contain_seg_end_point:
+                            end_ms = self.output_data_buf[i].end_ms
+                            self.next_seg = True
+                            self.output_data_buf_offset += 1
+                        else:
+                            end_ms = -1
+                            self.next_seg = False
+                    else:
+                        if not self.output_data_buf[i].contain_seg_start_point or not self.output_data_buf[
+                            i].contain_seg_end_point:
+                            continue
+                        start_ms = self.output_data_buf[i].start_ms
                         end_ms = self.output_data_buf[i].end_ms
-                        self.next_seg = True
                         self.output_data_buf_offset += 1
-                    else:
-                        end_ms = -1
-                        self.next_seg = False
                     segment = [start_ms, end_ms]
                     segment_batch.append(segment)
+
             if segment_batch:
                 segments.append(segment_batch)
         if is_final:
@@ -605,3 +613,4 @@ class E2EVadModel():
         if self.vad_state_machine == VadStateMachine.kVadInStateEndPointDetected and \
                 self.vad_opts.detect_mode == VadDetectMode.kVadMutipleUtteranceDetectMode.value:
             self.ResetDetection()
+

funasr/runtime/python/onnxruntime/funasr_onnx/utils/frontend.py

@@ -1,6 +1,7 @@
 # -*- encoding: utf-8 -*-
 from pathlib import Path
 from typing import Any, Dict, Iterable, List, NamedTuple, Set, Tuple, Union
+import copy
 
 import numpy as np
 from typeguard import check_argument_types
@@ -153,6 +154,187 @@ class WavFrontend():
         cmvn = np.array([means, vars])
         return cmvn
 
+
+class WavFrontendOnline(WavFrontend):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        # self.fbank_fn = knf.OnlineFbank(self.opts)
+        # add variables
+        self.frame_sample_length = int(self.opts.frame_opts.frame_length_ms * self.opts.frame_opts.samp_freq / 1000)
+        self.frame_shift_sample_length = int(self.opts.frame_opts.frame_shift_ms * self.opts.frame_opts.samp_freq / 1000)
+        self.waveform = None
+        self.reserve_waveforms = None
+        self.input_cache = None
+        self.lfr_splice_cache = []
+
+    @staticmethod
+    # inputs has catted the cache
+    def apply_lfr(inputs: np.ndarray, lfr_m: int, lfr_n: int, is_final: bool = False) -> Tuple[
+        np.ndarray, np.ndarray, int]:
+        """
+        Apply lfr with data
+        """
+
+        LFR_inputs = []
+        T = inputs.shape[0]  # include the right context
+        T_lfr = int(np.ceil((T - (lfr_m - 1) // 2) / lfr_n))  # minus the right context: (lfr_m - 1) // 2
+        splice_idx = T_lfr
+        for i in range(T_lfr):
+            if lfr_m <= T - i * lfr_n:
+                LFR_inputs.append((inputs[i * lfr_n:i * lfr_n + lfr_m]).reshape(1, -1))
+            else:  # process last LFR frame
+                if is_final:
+                    num_padding = lfr_m - (T - i * lfr_n)
+                    frame = (inputs[i * lfr_n:]).reshape(-1)
+                    for _ in range(num_padding):
+                        frame = np.hstack((frame, inputs[-1]))
+                    LFR_inputs.append(frame)
+                else:
+                    # update splice_idx and break the circle
+                    splice_idx = i
+                    break
+        splice_idx = min(T - 1, splice_idx * lfr_n)
+        lfr_splice_cache = inputs[splice_idx:, :]
+        LFR_outputs = np.vstack(LFR_inputs)
+        return LFR_outputs.astype(np.float32), lfr_splice_cache, splice_idx
+
+    @staticmethod
+    def compute_frame_num(sample_length: int, frame_sample_length: int, frame_shift_sample_length: int) -> int:
+        frame_num = int((sample_length - frame_sample_length) / frame_shift_sample_length + 1)
+        return frame_num if frame_num >= 1 and sample_length >= frame_sample_length else 0
+
+
+    def fbank(
+            self,
+            input: np.ndarray,
+            input_lengths: np.ndarray
+    ) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
+        self.fbank_fn = knf.OnlineFbank(self.opts)
+        batch_size = input.shape[0]
+        if self.input_cache is None:
+            self.input_cache = np.empty((batch_size, 0), dtype=np.float32)
+        input = np.concatenate((self.input_cache, input), axis=1)
+        frame_num = self.compute_frame_num(input.shape[-1], self.frame_sample_length, self.frame_shift_sample_length)
+        # update self.in_cache
+        self.input_cache = input[:, -(input.shape[-1] - frame_num * self.frame_shift_sample_length):]
+        waveforms = np.empty(0, dtype=np.int16)
+        feats_pad = np.empty(0, dtype=np.float32)
+        feats_lens = np.empty(0, dtype=np.int32)
+        if frame_num:
+            waveforms = []
+            feats = []
+            feats_lens = []
+            for i in range(batch_size):
+                waveform = input[i]
+                waveforms.append(
+                    waveform[:((frame_num - 1) * self.frame_shift_sample_length + self.frame_sample_length)])
+                waveform = waveform * (1 << 15)
+                
+                self.fbank_fn.accept_waveform(self.opts.frame_opts.samp_freq, waveform.tolist())
+                frames = self.fbank_fn.num_frames_ready
+                mat = np.empty([frames, self.opts.mel_opts.num_bins])
+                for i in range(frames):
+                    mat[i, :] = self.fbank_fn.get_frame(i)
+                feat = mat.astype(np.float32)
+                feat_len = np.array(mat.shape[0]).astype(np.int32)
+                feats.append(mat)
+                feats_lens.append(feat_len)
+
+            waveforms = np.stack(waveforms)
+            feats_lens = np.array(feats_lens)
+            feats_pad = np.array(feats)
+        self.fbanks = feats_pad
+        self.fbanks_lens = copy.deepcopy(feats_lens)
+        return waveforms, feats_pad, feats_lens
+
+    def get_fbank(self) -> Tuple[np.ndarray, np.ndarray]:
+        return self.fbanks, self.fbanks_lens
+
+    def lfr_cmvn(
+            self,
+            input: np.ndarray,
+            input_lengths: np.ndarray,
+            is_final: bool = False
+    ) -> Tuple[np.ndarray, np.ndarray, List[int]]:
+        batch_size = input.shape[0]
+        feats = []
+        feats_lens = []
+        lfr_splice_frame_idxs = []
+        for i in range(batch_size):
+            mat = input[i, :input_lengths[i], :]
+            lfr_splice_frame_idx = -1
+            if self.lfr_m != 1 or self.lfr_n != 1:
+                # update self.lfr_splice_cache in self.apply_lfr
+                mat, self.lfr_splice_cache[i], lfr_splice_frame_idx = self.apply_lfr(mat, self.lfr_m, self.lfr_n,
+                                                                                     is_final)
+            if self.cmvn_file is not None:
+                mat = self.apply_cmvn(mat)
+            feat_length = mat.shape[0]
+            feats.append(mat)
+            feats_lens.append(feat_length)
+            lfr_splice_frame_idxs.append(lfr_splice_frame_idx)
+
+        feats_lens = np.array(feats_lens)
+        feats_pad = np.array(feats)
+        return feats_pad, feats_lens, lfr_splice_frame_idxs
+
+
+    def extract_fbank(
+            self, input: np.ndarray, input_lengths: np.ndarray, is_final: bool = False
+    ) -> Tuple[np.ndarray, np.ndarray]:
+        batch_size = input.shape[0]
+        assert batch_size == 1, 'we support to extract feature online only when the batch size is equal to 1 now'
+        waveforms, feats, feats_lengths = self.fbank(input, input_lengths)  # input shape: B T D
+        if feats.shape[0]:
+            self.waveforms = waveforms if self.reserve_waveforms is None else np.concatenate(
+                (self.reserve_waveforms, waveforms), axis=1)
+            if not self.lfr_splice_cache:
+                for i in range(batch_size):
+                    self.lfr_splice_cache.append(np.expand_dims(feats[i][0, :], axis=0).repeat((self.lfr_m - 1) // 2, axis=0))
+            
+            if feats_lengths[0] + self.lfr_splice_cache[0].shape[0] >= self.lfr_m:
+                lfr_splice_cache_np = np.stack(self.lfr_splice_cache)  # B T D
+                feats = np.concatenate((lfr_splice_cache_np, feats), axis=1)
+                feats_lengths += lfr_splice_cache_np[0].shape[0]
+                frame_from_waveforms = int(
+                    (self.waveforms.shape[1] - self.frame_sample_length) / self.frame_shift_sample_length + 1)
+                minus_frame = (self.lfr_m - 1) // 2 if self.reserve_waveforms is None else 0
+                feats, feats_lengths, lfr_splice_frame_idxs = self.lfr_cmvn(feats, feats_lengths, is_final)
+                if self.lfr_m == 1:
+                    self.reserve_waveforms = None
+                else:
+                    reserve_frame_idx = lfr_splice_frame_idxs[0] - minus_frame
+                    # print('reserve_frame_idx:  ' + str(reserve_frame_idx))
+                    # print('frame_frame:  ' + str(frame_from_waveforms))
+                    self.reserve_waveforms = self.waveforms[:, reserve_frame_idx * self.frame_shift_sample_length:frame_from_waveforms * self.frame_shift_sample_length]
+                    sample_length = (frame_from_waveforms - 1) * self.frame_shift_sample_length + self.frame_sample_length
+                    self.waveforms = self.waveforms[:, :sample_length]
+            else:
+                # update self.reserve_waveforms and self.lfr_splice_cache
+                self.reserve_waveforms = self.waveforms[:,
+                                         :-(self.frame_sample_length - self.frame_shift_sample_length)]
+                for i in range(batch_size):
+                    self.lfr_splice_cache[i] = np.concatenate((self.lfr_splice_cache[i], feats[i]), axis=0)
+                return np.empty(0, dtype=np.float32), feats_lengths
+        else:
+            if is_final:
+                self.waveforms = waveforms if self.reserve_waveforms is None else self.reserve_waveforms
+                feats = np.stack(self.lfr_splice_cache)
+                feats_lengths = np.zeros(batch_size, dtype=np.int32) + feats.shape[1]
+                feats, feats_lengths, _ = self.lfr_cmvn(feats, feats_lengths, is_final)
+        if is_final:
+            self.cache_reset()
+        return feats, feats_lengths
+
+    def get_waveforms(self):
+        return self.waveforms
+
+    def cache_reset(self):
+        self.fbank_fn = knf.OnlineFbank(self.opts)
+        self.reserve_waveforms = None
+        self.input_cache = None
+        self.lfr_splice_cache = []
+
 def load_bytes(input):
     middle_data = np.frombuffer(input, dtype=np.int16)
     middle_data = np.asarray(middle_data)
@@ -188,4 +370,4 @@ def test():
     return feat, feat_len
 
 if __name__ == '__main__':
-    test()
+    test()

funasr/runtime/python/onnxruntime/funasr_onnx/vad_bin.py

@@ -59,37 +59,48 @@ class Fsmn_vad():
 		
 	
 	def __call__(self, audio_in: Union[str, np.ndarray, List[str]], **kwargs) -> List:
-		# waveform_list = self.load_data(audio_in, self.frontend.opts.frame_opts.samp_freq)
-		
-		param_dict = kwargs.get('param_dict', dict())
-		is_final = param_dict.get('is_final', False)
-		audio_in_cache = param_dict.get('audio_in_cache', None)
-		audio_in_cum = audio_in
-		if audio_in_cache is not None:
-			audio_in_cum = np.concatenate((audio_in_cache, audio_in_cum))
-		param_dict['audio_in_cache'] = audio_in_cum
-		feats, feats_len = self.extract_feat([audio_in_cum])
-		
-		in_cache = param_dict.get('in_cache', list())
-		in_cache = self.prepare_cache(in_cache)
-		beg_idx = param_dict.get('beg_idx',0)
-		feats = feats[:, beg_idx:beg_idx+8, :]
-		param_dict['beg_idx'] = beg_idx + feats.shape[1]
-		try:
-			inputs = [feats]
-			inputs.extend(in_cache)
-			scores, out_caches = self.infer(inputs)
-			param_dict['in_cache'] = out_caches
-			segments = self.vad_scorer(scores, audio_in[None, :], is_final=is_final, max_end_sil=self.max_end_sil)
-			# print(segments)
-			if len(segments) == 1 and segments[0][0][1] != -1:
-				self.frontend.reset_status()
-			
+		waveform_list = self.load_data(audio_in, self.frontend.opts.frame_opts.samp_freq)
+		waveform_nums = len(waveform_list)
+		is_final = kwargs.get('kwargs', False)
+
+		segments = [[]] * self.batch_size
+		for beg_idx in range(0, waveform_nums, self.batch_size):
 			
-		except ONNXRuntimeError:
-			logging.warning(traceback.format_exc())
-			logging.warning("input wav is silence or noise")
-			segments = []
+			end_idx = min(waveform_nums, beg_idx + self.batch_size)
+			waveform = waveform_list[beg_idx:end_idx]
+			feats, feats_len = self.extract_feat(waveform)
+			waveform = np.array(waveform)
+			param_dict = kwargs.get('param_dict', dict())
+			in_cache = param_dict.get('in_cache', list())
+			in_cache = self.prepare_cache(in_cache)
+			try:
+				t_offset = 0
+				step = int(min(feats_len.max(), 6000))
+				for t_offset in range(0, int(feats_len), min(step, feats_len - t_offset)):
+					if t_offset + step >= feats_len - 1:
+						step = feats_len - t_offset
+						is_final = True
+					else:
+						is_final = False
+					feats_package = feats[:, t_offset:int(t_offset + step), :]
+					waveform_package = waveform[:, t_offset * 160:min(waveform.shape[-1], (int(t_offset + step) - 1) * 160 + 400)]
+
+					inputs = [feats_package]
+					# inputs = [feats]
+					inputs.extend(in_cache)
+					scores, out_caches = self.infer(inputs)
+					in_cache = out_caches
+					segments_part = self.vad_scorer(scores, waveform_package, is_final=is_final, max_end_sil=self.max_end_sil, online=False)
+					# segments = self.vad_scorer(scores, waveform[0][None, :], is_final=is_final, max_end_sil=self.max_end_sil)
+
+					if segments_part:
+						for batch_num in range(0, self.batch_size):
+							segments[batch_num] += segments_part[batch_num]
+				
+			except ONNXRuntimeError:
+				# logging.warning(traceback.format_exc())
+				logging.warning("input wav is silence or noise")
+				segments = ''
 	
 		return segments
 
@@ -140,4 +151,4 @@ class Fsmn_vad():
 		outputs = self.ort_infer(feats)
 		scores, out_caches = outputs[0], outputs[1:]
 		return scores, out_caches
-	
+	

funasr/runtime/python/onnxruntime/funasr_onnx/vad_online_bin.py

@@ -0,0 +1,134 @@
+# -*- encoding: utf-8 -*-
+
+import os.path
+from pathlib import Path
+from typing import List, Union, Tuple
+
+import copy
+import librosa
+import numpy as np
+
+from .utils.utils import (ONNXRuntimeError,
+                          OrtInferSession, get_logger,
+                          read_yaml)
+from .utils.frontend import WavFrontendOnline
+from .utils.e2e_vad import E2EVadModel
+
+logging = get_logger()
+
+
+class Fsmn_vad():
+	def __init__(self, model_dir: Union[str, Path] = None,
+	             batch_size: int = 1,
+	             device_id: Union[str, int] = "-1",
+	             quantize: bool = False,
+	             intra_op_num_threads: int = 4,
+	             max_end_sil: int = None,
+	             ):
+		
+		if not Path(model_dir).exists():
+			raise FileNotFoundError(f'{model_dir} does not exist.')
+		
+		model_file = os.path.join(model_dir, 'model.onnx')
+		if quantize:
+			model_file = os.path.join(model_dir, 'model_quant.onnx')
+		config_file = os.path.join(model_dir, 'vad.yaml')
+		cmvn_file = os.path.join(model_dir, 'vad.mvn')
+		config = read_yaml(config_file)
+		
+		self.frontend = WavFrontendOnline(
+			cmvn_file=cmvn_file,
+			**config['frontend_conf']
+		)
+		self.ort_infer = OrtInferSession(model_file, device_id, intra_op_num_threads=intra_op_num_threads)
+		self.batch_size = batch_size
+		self.vad_scorer = E2EVadModel(config["vad_post_conf"])
+		self.max_end_sil = max_end_sil if max_end_sil is not None else config["vad_post_conf"]["max_end_silence_time"]
+		self.encoder_conf = config["encoder_conf"]
+	
+	def prepare_cache(self, in_cache: list = []):
+		if len(in_cache) > 0:
+			return in_cache
+		fsmn_layers = self.encoder_conf["fsmn_layers"]
+		proj_dim = self.encoder_conf["proj_dim"]
+		lorder = self.encoder_conf["lorder"]
+		for i in range(fsmn_layers):
+			cache = np.zeros((1, proj_dim, lorder-1, 1)).astype(np.float32)
+			in_cache.append(cache)
+		return in_cache
+		
+	
+	def __call__(self, audio_in: np.ndarray, **kwargs) -> List:
+		waveforms = np.expand_dims(audio_in, axis=0)
+		
+		param_dict = kwargs.get('param_dict', dict())
+		is_final = param_dict.get('is_final', False)
+		feats, feats_len = self.extract_feat(waveforms, is_final)
+		segments = []
+		if feats.size != 0:
+			in_cache = param_dict.get('in_cache', list())
+			in_cache = self.prepare_cache(in_cache)
+			try:
+				inputs = [feats]
+				inputs.extend(in_cache)
+				scores, out_caches = self.infer(inputs)
+				param_dict['in_cache'] = out_caches
+				waveforms = self.frontend.get_waveforms()
+				segments = self.vad_scorer(scores, waveforms, is_final=is_final, max_end_sil=self.max_end_sil, online=True)
+
+
+			except ONNXRuntimeError:
+				logging.warning(traceback.format_exc())
+				logging.warning("input wav is silence or noise")
+				segments = []
+		return segments
+
+	def load_data(self,
+	              wav_content: Union[str, np.ndarray, List[str]], fs: int = None) -> List:
+		def load_wav(path: str) -> np.ndarray:
+			waveform, _ = librosa.load(path, sr=fs)
+			return waveform
+		
+		if isinstance(wav_content, np.ndarray):
+			return [wav_content]
+		
+		if isinstance(wav_content, str):
+			return [load_wav(wav_content)]
+		
+		if isinstance(wav_content, list):
+			return [load_wav(path) for path in wav_content]
+		
+		raise TypeError(
+			f'The type of {wav_content} is not in [str, np.ndarray, list]')
+	
+	def extract_feat(self,
+	                 waveforms: np.ndarray, is_final: bool = False
+	                 ) -> Tuple[np.ndarray, np.ndarray]:
+		waveforms_lens = np.zeros(waveforms.shape[0]).astype(np.int32)
+		for idx, waveform in enumerate(waveforms):
+			waveforms_lens[idx] = waveform.shape[-1]
+
+		feats, feats_len = self.frontend.extract_fbank(waveforms, waveforms_lens, is_final)
+		# feats.append(feat)
+		# feats_len.append(feat_len)
+
+		# feats = self.pad_feats(feats, np.max(feats_len))
+		# feats_len = np.array(feats_len).astype(np.int32)
+		return feats.astype(np.float32), feats_len.astype(np.int32)
+
+	@staticmethod
+	def pad_feats(feats: List[np.ndarray], max_feat_len: int) -> np.ndarray:
+		def pad_feat(feat: np.ndarray, cur_len: int) -> np.ndarray:
+			pad_width = ((0, max_feat_len - cur_len), (0, 0))
+			return np.pad(feat, pad_width, 'constant', constant_values=0)
+		
+		feat_res = [pad_feat(feat, feat.shape[0]) for feat in feats]
+		feats = np.array(feat_res).astype(np.float32)
+		return feats
+	
+	def infer(self, feats: List) -> Tuple[np.ndarray, np.ndarray]:
+		
+		outputs = self.ort_infer(feats)
+		scores, out_caches = outputs[0], outputs[1:]
+		return scores, out_caches
+