added buffer support
This commit is contained in:
parent
e2581c8aad
commit
a0337b36a1
16
setup.py
16
setup.py
|
@ -8,20 +8,20 @@ if sys.platform == 'darwin':
|
|||
os.environ['CXXFLAGS'] = '-DGGML_USE_ACCELERATE -O3 -std=c++11'
|
||||
os.environ['LDFLAGS'] = '-framework Accelerate'
|
||||
else:
|
||||
os.environ['CFLAGS'] = '-mavx -mavx2 -mfma -mf16c -O3 -std=gnu11'
|
||||
os.environ['CXXFLAGS'] = '-mavx -mavx2 -mfma -mf16c -O3 -std=c++11'
|
||||
os.environ['CFLAGS'] = '-mavx -mavx2 -mfma -mf16c -O3 /std:c++14'
|
||||
os.environ['CXXFLAGS'] = '-mavx -mavx2 -mfma -mf16c -O3 /std:c++14'
|
||||
|
||||
ext_modules = [
|
||||
Extension(
|
||||
name="whispercpp",
|
||||
sources=["whispercpp.pyx", "whisper.cpp/whisper.cpp"],
|
||||
sources=["C:\\Users\\aidan\\Desktop\\Code\\projectMesa\\whispercpp.py\\whispercpp.pyx", "C:\\Users\\aidan\\Desktop\\Code\\projectMesa\\whispercpp.py\\whisper.cpp/whisper.cpp"],
|
||||
language="c++",
|
||||
extra_compile_args=["-std=c++11"],
|
||||
)
|
||||
extra_compile_args=["/std:c++14"],
|
||||
)
|
||||
]
|
||||
ext_modules = cythonize(ext_modules)
|
||||
|
||||
whisper_clib = ('whisper_clib', {'sources': ['whisper.cpp/ggml.c']})
|
||||
whisper_clib = ('whisper_clib', {'sources': ['C:\\Users\\aidan\\Desktop\\Code\\projectMesa\\whispercpp.py\\whisper.cpp/ggml.c']})
|
||||
|
||||
setup(
|
||||
name='whispercpp',
|
||||
|
@ -30,8 +30,8 @@ setup(
|
|||
author='Luke Southam',
|
||||
author_email='luke@devthe.com',
|
||||
libraries=[whisper_clib],
|
||||
ext_modules = cythonize("whispercpp.pyx"),
|
||||
include_dirs = ['./whisper.cpp/', numpy.get_include()],
|
||||
ext_modules=cythonize(ext_modules),
|
||||
include_dirs=['C:\\Users\\aidan\\Desktop\\Code\\projectMesa\\whispercpp.py\\whisper.cpp', numpy.get_include()],
|
||||
install_requires=[
|
||||
'numpy',
|
||||
'ffmpeg-python',
|
||||
|
|
|
@ -1 +1 @@
|
|||
Subproject commit 1d716d6e34f3f4ba57bd9706a9258a0bdb008153
|
||||
Subproject commit 0a2d1210bcb98978214bbf4e100922a413afd39d
|
File diff suppressed because it is too large
Load Diff
|
@ -71,7 +71,8 @@ cdef extern from "whisper.h" nogil:
|
|||
whisper_encoder_begin_callback encoder_begin_callback
|
||||
void* encoder_begin_callback_user_data
|
||||
whisper_full_params whisper_full_default_params(whisper_sampling_strategy)
|
||||
cdef whisper_context* whisper_init(char*)
|
||||
cdef whisper_context* whisper_init_from_file(char*)
|
||||
cdef whisper_context* whisper_init_from_buffer(voidptr, int)
|
||||
cdef void whisper_free(whisper_context*)
|
||||
cdef int whisper_pcm_to_mel(whisper_context*, float*, int, int)
|
||||
cdef int whisper_set_mel(whisper_context*, float*, int, int)
|
||||
|
|
|
@ -16,7 +16,7 @@ cimport numpy as cnp
|
|||
cdef int SAMPLE_RATE = 16000
|
||||
cdef char* TEST_FILE = 'test.wav'
|
||||
cdef char* DEFAULT_MODEL = 'tiny'
|
||||
cdef char* LANGUAGE = b'fr'
|
||||
cdef char* LANGUAGE = b'en'
|
||||
cdef int N_THREADS = os.cpu_count()
|
||||
|
||||
MODELS = {
|
||||
|
@ -84,21 +84,37 @@ cdef class Whisper:
|
|||
cdef whisper_context * ctx
|
||||
cdef whisper_full_params params
|
||||
|
||||
def __init__(self, model=DEFAULT_MODEL, pb=None):
|
||||
model_fullname = f'ggml-{model}.bin'.encode('utf8')
|
||||
def __init__(self, model=DEFAULT_MODEL, pb=None, buf=None):
|
||||
|
||||
model_fullname = f'ggml-{model}.bin'
|
||||
download_model(model_fullname)
|
||||
model_path = Path(MODELS_DIR).joinpath(model_fullname)
|
||||
cdef bytes model_b = str(model_path).encode('utf8')
|
||||
self.ctx = whisper_init(model_b)
|
||||
|
||||
if buf is not None:
|
||||
self.ctx = whisper_init_from_buffer(buf, buf.size)
|
||||
else:
|
||||
self.ctx = whisper_init_from_file(model_b)
|
||||
|
||||
self.params = default_params()
|
||||
whisper_print_system_info()
|
||||
|
||||
|
||||
def __dealloc__(self):
|
||||
whisper_free(self.ctx)
|
||||
|
||||
def transcribe(self, filename=TEST_FILE):
|
||||
print("Loading data..")
|
||||
cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] frames = load_audio(<bytes>filename)
|
||||
if (type(filename) == np.ndarray) :
|
||||
temp = filename
|
||||
|
||||
elif (type(filename) == str) :
|
||||
temp = load_audio(<bytes>filename)
|
||||
else :
|
||||
temp = load_audio(<bytes>TEST_FILE)
|
||||
|
||||
|
||||
cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] frames = temp
|
||||
|
||||
print("Transcribing..")
|
||||
return whisper_full(self.ctx, self.params, &frames[0], len(frames))
|
||||
|
|
Loading…
Reference in New Issue