71 lines
2.5 KiB
Python
71 lines
2.5 KiB
Python
import io
|
|
from pydub import AudioSegment
|
|
from pydub.silence import split_on_silence
|
|
|
|
def segment_audio(audio_data: bytes, frame_rate: int, channels: int, max_length_s: int, silence_thresh_db: int, min_silence_len_ms: int = 500, keep_silence_ms: int = 250):
|
|
"""
|
|
将音频根据静音和最大长度进行分割。
|
|
新逻辑：优先在静音处分割，然后将短片段合并，使其尽可能接近但不超过最大长度。
|
|
|
|
:param audio_data: 原始音频数据 (bytes).
|
|
:param frame_rate: 音频的采样率.
|
|
:param channels: 音频的通道数.
|
|
:param max_length_s: 每个片段的最大长度（秒）.
|
|
:param silence_thresh_db: 低于此分贝值被认为是静音.
|
|
:param min_silence_len_ms: 最小静音长度（毫秒），用于分割.
|
|
:param keep_silence_ms: 在分割后的片段开头和结尾保留的静音长度（毫秒）.
|
|
:return: 一个包含 pydub.AudioSegment 对象的列表.
|
|
"""
|
|
if not audio_data:
|
|
raise ValueError("输入的音频数据不能为空。")
|
|
|
|
# 从原始字节创建 pydub AudioSegment
|
|
# sample_width=2 对应 16-bit PCM
|
|
audio = AudioSegment(
|
|
data=audio_data,
|
|
sample_width=2,
|
|
frame_rate=frame_rate,
|
|
channels=channels
|
|
)
|
|
|
|
max_length_ms = max_length_s * 1000
|
|
|
|
# 1. 首先根据静音进行分割
|
|
initial_chunks = split_on_silence(
|
|
audio,
|
|
min_silence_len=min_silence_len_ms,
|
|
silence_thresh=silence_thresh_db,
|
|
keep_silence=keep_silence_ms
|
|
)
|
|
|
|
if not initial_chunks:
|
|
initial_chunks = [audio]
|
|
|
|
# 2. 合并短片段，使其长度接近 max_length_ms
|
|
combined_chunks = []
|
|
current_chunk = None
|
|
|
|
for chunk in initial_chunks:
|
|
if current_chunk is None:
|
|
current_chunk = chunk
|
|
elif len(current_chunk) + len(chunk) <= max_length_ms:
|
|
current_chunk += chunk
|
|
else:
|
|
combined_chunks.append(current_chunk)
|
|
current_chunk = chunk
|
|
|
|
if current_chunk is not None:
|
|
combined_chunks.append(current_chunk)
|
|
|
|
# 3. 再次检查是否有片段超过最大长度（如果单个静音分割后的片段就超长了）
|
|
# 并将其暴力分割
|
|
final_chunks = []
|
|
for chunk in combined_chunks:
|
|
if len(chunk) > max_length_ms:
|
|
for i in range(0, len(chunk), max_length_ms):
|
|
final_chunks.append(chunk[i:i + max_length_ms])
|
|
else:
|
|
final_chunks.append(chunk)
|
|
|
|
return final_chunks
|