ii-vision/transcoder/audio.py

101 lines
3.0 KiB
Python

"""Encodes input audio stream into sequence of speaker duty cycle counts."""
from typing import Iterator
import audioread
import librosa
import numpy as np
class Audio:
"""
Decodes audio stream from input file and resamples.
Notes on audio bitrate:
At 73 cycles/tick, true audio playback sample rate is
roughly 1024*1024/73 = 14364 Hz (ignoring ACK slow path).
Typical audio encoding is 44100Hz which is close to 14700*3
Downscaling by 3x gives better results than trying to resample
to a non-divisor. So we cheat a bit and play back the video a
tiny bit (<2%) faster.
For //gs playback at 2.8MHz, the effective speed increase is only about
1.6x. This is probably because accessing the I/O page is done at 1MHz
to not mess up hardware timings.
This is close (2.1%) to 22500Hz which is again a simple divisor of the
base frequency (1/2).
"""
def __init__(
self,
filename: str,
bitrate: int = 14700,
normalization: float = None):
self.filename = filename # type: str
# TODO: take into account that the available range is slightly offset
# as fraction of total cycle count?
self._tick_range = [4, 66]
self.sample_rate = float(bitrate) # type: float
self.normalization = (
normalization or self._normalization()) # type: float
def _decode(self, f, buf) -> np.array:
"""
:param f:
:param buf:
:return:
"""
data = np.frombuffer(buf, dtype='int16').astype(
'float32').reshape((f.channels, -1), order='F')
a = librosa.core.to_mono(data)
a = librosa.resample(a, orig_sr=f.samplerate,
target_sr=self.sample_rate,
res_type='scipy', scale=True).flatten()
return a
def _normalization(self, read_bytes=1024 * 1024 * 10):
"""Read first read_bytes of audio stream and compute normalization.
We normalize based on the 0.5th and 99.5th percentiles, i.e. only <1% of
samples will clip.
:param read_bytes:
:return:
"""
raw = bytearray()
with audioread.audio_open(self.filename) as f:
for buf in f.read_data():
raw.extend(bytearray(buf))
if len(raw) > read_bytes:
break
a = self._decode(f, raw)
norm = np.max(np.abs(np.percentile(a, [0.5, 99.5])))
return 16384. / norm
def audio_stream(self) -> Iterator[int]:
"""
:return:
"""
with audioread.audio_open(self.filename) as f:
for buf in f.read_data(128 * 1024):
a = self._decode(f, buf)
a /= 16384 # normalize to -1.0 .. 1.0
a *= self.normalization
# Convert to -16 .. 16
a = (a * 16).astype(np.int)
a = np.clip(a, -15, 16)
yield from a