From 3f8294b9d7deaa0cbdb0b2ef6f3802d46ae133a9 Mon Sep 17 00:00:00 2001
From: zhifu gao <zhifu.gzf@alibaba-inc.com>
Date: 星期三, 25 十二月 2024 17:16:11 +0800
Subject: [PATCH] Revert "shfit to shift (#2266)" (#2336)
---
funasr/utils/load_utils.py | 45 +++++++++++++++++++++++++++++++++++++++++++--
1 files changed, 43 insertions(+), 2 deletions(-)
diff --git a/funasr/utils/load_utils.py b/funasr/utils/load_utils.py
index faafc0f..1d80fcf 100644
--- a/funasr/utils/load_utils.py
+++ b/funasr/utils/load_utils.py
@@ -1,6 +1,7 @@
import os
import torch
import json
+from io import BytesIO
import torch.distributed as dist
import numpy as np
import kaldiio
@@ -17,6 +18,11 @@
import pdb
import subprocess
from subprocess import CalledProcessError, run
+
+try:
+ from pydub import AudioSegment
+except:
+ pass
def is_ffmpeg_installed():
@@ -76,7 +82,7 @@
for audio in data_or_path_or_list
]
if isinstance(data_or_path_or_list, str) and data_or_path_or_list.startswith(
- ("http://", "https://")
+ ("http://", "https://")
): # download url to local file
data_or_path_or_list = download_from_url(data_or_path_or_list)
@@ -112,7 +118,7 @@
elif isinstance(data_or_path_or_list, str) and data_type == "text" and tokenizer is not None:
data_or_path_or_list = tokenizer.encode(data_or_path_or_list)
elif isinstance(data_or_path_or_list, np.ndarray): # audio sample point
- data_or_path_or_list = torch.from_numpy(data_or_path_or_list).squeeze() # [n_samples,]
+ data_or_path_or_list = torch.from_numpy(data_or_path_or_list) # .squeeze() # [n_samples,]
elif isinstance(data_or_path_or_list, str) and data_type == "kaldi_ark":
data_mat = kaldiio.load_mat(data_or_path_or_list)
if isinstance(data_mat, tuple):
@@ -136,6 +142,10 @@
def load_bytes(input):
+ try:
+ input = validate_frame_rate(input)
+ except:
+ pass
middle_data = np.frombuffer(input, dtype=np.int16)
middle_data = np.asarray(middle_data)
if middle_data.dtype.kind not in "iu":
@@ -151,6 +161,37 @@
return array
+def validate_frame_rate(
+ input,
+ fs: int = 16000,
+):
+
+ # 灏嗘枃浠惰鍙栦负瀛楄妭娴�
+ byte_data = BytesIO(input)
+
+ # 浣跨敤 pydub 鍔犺浇闊抽
+ try:
+ audio = AudioSegment.from_file(byte_data)
+ except:
+ raise RuntimeError(
+ "You are decoding the pcm data, please install pydub first. via `pip install pydub`."
+ )
+
+ # 纭繚閲囨牱鐜囦负 16000 Hz
+ if audio.frame_rate != fs:
+ audio = audio.set_frame_rate(fs)
+
+ # 灏嗛噸鏂伴噰鏍峰悗鐨勯煶棰戝鍑轰负瀛楄妭娴�
+ output = BytesIO()
+ audio.export(output, format="wav")
+ output.seek(0)
+
+ # 鑾峰彇閲嶆柊閲囨牱鍚庣殑瀛楄妭娴佹暟鎹�
+ input = output.read()
+
+ return input
+
+
def extract_fbank(data, data_len=None, data_type: str = "sound", frontend=None, **kwargs):
if isinstance(data, np.ndarray):
data = torch.from_numpy(data)
--
Gitblit v1.9.1