1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
|
"""Pure streaming voice-clip state machine."""
from __future__ import annotations
from collections import deque
from dataclasses import dataclass
from datetime import datetime, timedelta
import numpy as np
SAMPLE_RATE = 16_000
@dataclass(frozen=True)
class Start:
started_at: datetime
samples: np.ndarray
@dataclass(frozen=True)
class Audio:
samples: np.ndarray
@dataclass(frozen=True)
class End:
pass
class Segmenter:
"""Emit stream events while retaining only the configured pre-roll in memory."""
def __init__(
self,
*,
pre_roll_seconds: float,
silence_seconds: float,
max_segment_seconds: float = 600,
sample_rate: int = SAMPLE_RATE,
) -> None:
self.sample_rate = sample_rate
self.pre_roll_samples = round(pre_roll_seconds * sample_rate)
self.silence_samples_limit = round(silence_seconds * sample_rate)
self.max_segment_samples = round(max_segment_seconds * sample_rate)
self._pre_roll: deque[np.ndarray] = deque()
self._pre_roll_size = 0
self._active = False
self._segment_samples = 0
self._silence_samples = 0
@property
def active(self) -> bool:
return self._active
def push(
self, samples: np.ndarray, is_speech: bool, captured_at: datetime
) -> tuple[Start | Audio | End, ...]:
samples = np.asarray(samples, dtype=np.float32)
if samples.ndim != 1:
raise ValueError("audio must be a mono, one-dimensional array")
if not len(samples):
return ()
if not self._active:
if not is_speech:
self._append_pre_roll(samples)
return ()
pre_roll = self._take_pre_roll()
started_at = captured_at - timedelta(
seconds=len(pre_roll) / self.sample_rate
)
self._active = True
self._segment_samples = len(pre_roll) + len(samples)
self._silence_samples = 0
self._append_pre_roll(samples)
return (Start(started_at, np.concatenate((pre_roll, samples))),)
if self._segment_samples + len(samples) > self.max_segment_samples:
self._append_pre_roll(samples)
self._active = True
self._segment_samples = len(samples)
self._silence_samples = 0
return (End(), Start(captured_at, samples))
self._segment_samples += len(samples)
if is_speech:
self._silence_samples = 0
else:
self._silence_samples += len(samples)
self._append_pre_roll(samples)
events: tuple[Start | Audio | End, ...] = (Audio(samples),)
if self._silence_samples >= self.silence_samples_limit:
self._active = False
self._segment_samples = 0
self._silence_samples = 0
events += (End(),)
return events
def _append_pre_roll(self, samples: np.ndarray) -> None:
self._pre_roll.append(samples)
self._pre_roll_size += len(samples)
while self._pre_roll_size > self.pre_roll_samples:
excess = self._pre_roll_size - self.pre_roll_samples
oldest = self._pre_roll[0]
if len(oldest) <= excess:
self._pre_roll.popleft()
self._pre_roll_size -= len(oldest)
else:
self._pre_roll[0] = oldest[excess:]
self._pre_roll_size -= excess
def _take_pre_roll(self) -> np.ndarray:
if not self._pre_roll:
return np.empty(0, dtype=np.float32)
samples = np.concatenate(tuple(self._pre_roll))
self._pre_roll.clear()
self._pre_roll_size = 0
return samples
|
