Fox's Git Mirrors
pycodec2/pycodec2.pyx HEAD (438ee4f2) Text, 4.35 KB
#cython: language_level=3
# The language level specification is needed. I don't have it in my build file,
# so I use it per-module.
from .codec2 cimport *
import math
import numpy as np
cimport numpy as cnp
ctypedef cnp.int8_t CHAR_DTYPE_t
ctypedef cnp.int16_t SHORT_DTYPE_t
cnp.import_array()
_modes = {
700 : CODEC2_MODE_700C,
1200 : CODEC2_MODE_1200,
1300 : CODEC2_MODE_1300,
1400 : CODEC2_MODE_1400,
1600 : CODEC2_MODE_1600,
2400 : CODEC2_MODE_2400,
3200 : CODEC2_MODE_3200,
}
cdef class Codec2:
'''Wrapper for codec2 state and its functions.
Initialization method expects an integer defining expected bitrate per
second.'''
cdef CODEC2 *_c_codec2_state
def __cinit__(self, mode):
self._c_codec2_state = codec2_create(_modes[mode])
if self._c_codec2_state is NULL:
raise MemoryError()
def __dealloc__(self):
if self._c_codec2_state is not NULL:
codec2_destroy(self._c_codec2_state)
def encode(self, cnp.ndarray[SHORT_DTYPE_t, ndim=1] speech_in):
'''Encode samples with codec2.
Encode the given ndarray of samples to bits represented as a byte array.'''
assert len(speech_in) % self.samples_per_frame() == 0
frames = len(speech_in) // self.samples_per_frame()
bit_count = frames * self.bits_per_frame()
bits = b'\\x00' * int(math.ceil(bit_count / 8.0))
codec2_encode(self._c_codec2_state, bits, <short*>cnp.PyArray_DATA(speech_in))
return bits
def decode(self, bytes frames):
'''Decode a byte array into an ndarray of samples'''
assert len(frames) >= self.bytes_per_frame()
cdef cnp.ndarray[SHORT_DTYPE_t, ndim=1] speech_out
frame_count = int(math.floor(len(frames) / self.bytes_per_frame()))
sample_count = frame_count * self.samples_per_frame()
speech_out = np.empty(sample_count, dtype=np.int16, order='C')
codec2_decode(self._c_codec2_state, <short *>(cnp.PyArray_DATA(speech_out)), frames)
return speech_out
def decode_ber(self, bytes frames, float ber_est):
assert len(frames) >= self.bytes_per_frame()
cdef cnp.ndarray[SHORT_DTYPE_t, ndim=1] speech_out
frame_count = len(frames) // self.bytes_per_frame()
sample_count = frame_count * self.samples_per_frame()
speech_out = np.empty(sample_count, dtype=np.int16, order='C')
codec2_decode_ber(self._c_codec2_state,
<short *>(cnp.PyArray_DATA(speech_out)),
frames,
ber_est)
return speech_out
def samples_per_frame(self):
return codec2_samples_per_frame(self._c_codec2_state)
def bits_per_frame(self):
return codec2_bits_per_frame(self._c_codec2_state)
def bytes_per_frame(self):
return codec2_bytes_per_frame(self._c_codec2_state)
def set_lpc_post_filter(self,
int enable,
int bass_boost,
float beta,
float gamma):
codec2_set_lpc_post_filter(self._c_codec2_state,
enable,
bass_boost,
beta,
gamma)
def get_spare_bit_index(self):
return codec2_get_spare_bit_index(self._c_codec2_state)
def rebuild_spare_bit(self, cnp.ndarray[char, ndim=1] unpacked_bits):
return codec2_rebuild_spare_bit(self._c_codec2_state, <char *>cnp.PyArray_DATA(unpacked_bits))
def set_natural_or_gray(self, int gray):
codec2_set_natural_or_gray(self._c_codec2_state, gray)
def set_softdec(self, float softdec):
codec2_set_softdec(self._c_codec2_state, &softdec)
def get_energy(self, cnp.ndarray[unsigned char, ndim=1] bits):
return codec2_get_energy(self._c_codec2_state, <unsigned char*>(cnp.PyArray_DATA(bits)))
# support for ML and VQ experiments
def open_mlfeat(self, bytearray feat_filename, bytearray model_filename):
codec2_open_mlfeat(self._c_codec2_state, feat_filename, model_filename)
def load_codebook(self, bytearray feat_filename, int num, bytearray filename):
codec2_load_codebook(self._c_codec2_state, num, filename)
def get_var(self):
return codec2_get_var(self._c_codec2_state)
def enable_user_ratek(self):
cdef int K
cdef float *user_rate_k = codec2_enable_user_ratek(self._c_codec2_state, &K)
return (user_rate_k[0], K)
# 700C post filter and equaliser
def post_filter_700c(self, int en):
codec2_700c_post_filter(self._c_codec2_state, en)
def eq_700c(self, int en):
codec2_700c_eq(self._c_codec2_state, en)
Served by rngit 1.4.1 - Generated in 0.02s