init

Changes made (not exhaustive):
- changed defaults
- whisper.cpp submodule set to 1.2.0
- removed `requests` dependency
- models dir can be changed in constructor
- added support for setting params
- added back support for `large-v1` model
- added support for english-only models

.github/workflows/build_wheels.yml

@@ -1,40 +0,0 @@
-name: build_wheels
-run-name: ${{ github.actor }} is building wheels
-on: [push]
-jobs:
-  build_wheels:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-    steps:
-      - uses: actions/checkout@v2
-      - name: Checkout submodules
-        run: |
-          git submodule update --init --recursive
-  
-      - uses: actions/setup-python@v2
-        with:
-          python-version: '3.10'
-
-      - name: Setup pip
-        run: |
-           python -m pip install --upgrade pip
-           python -m pip install cibuildwheel==1.6.4
-         
-      - name:  Install
-        run:   |
-               if [ "$RUNNER_OS" == "Linux" ]; then
-                    sudo apt-get update
-                    sudo apt-get install gcc g++
-               fi
-        shell: bash
-
-      - name: Build wheel
-        run: python -m cibuildwheel --output-dir dist/
-        env:
-          CIBW_BUILD: cp36-* cp37-* cp38-*
-
-      - uses: actions/upload-artifact@v2
-        with:
-          path: ./dist/*.whl

.gitignore

@@ -127,3 +127,5 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+
+whispercpp.cpp

.gitmodules

@@ -1,3 +1,4 @@
 [submodule "whisper.cpp"]
 	path = whisper.cpp
 	url = https://github.com/ggerganov/whisper.cpp
+	branch = b2083c5d02db9a1e6dbb3d58254fd65ebfff4b5d

README.md

@@ -1,12 +1,23 @@
 Python bindings for whisper.cpp
 ===============================
 
-`pip install git+https://github.com/o4dev/whispercpp.py`
+```
+git clone --recurse-submodules https://git.ecker.tech/lightmare/whispercpp.py
+cd whispercpp.py
+pip install .
+```
+or
+```
+git clone https://git.ecker.tech/lightmare/whispercpp.py
+cd whispercpp.py
+git submodule update --init
+pip install .
+```
 
 ```python
 from whispercpp import Whisper
 
-w = Whisper('tiny')
+w = Whisper('tiny', models_dir='./models/', language=b'en')
 
 result = w.transcribe("myfile.mp3")
 text = w.extract_text(result)
@@ -14,3 +25,12 @@ text = w.extract_text(result)
 
 Note: default parameters might need to be tweaked.  
 See Whispercpp.pyx.
+
+Changes made (not exhaustive):
+- changed defaults
+- whisper.cpp submodule set to 1.2.0
+- removed `requests` dependency
+- models dir can be changed in constructor
+- added support for setting params
+- added back support for `large-v1` model
+- added support for english-only models

requirements.txt

@@ -0,0 +1,3 @@
+Cython
+numpy
+ffmpeg-python

setup.py

@@ -34,7 +34,6 @@ setup(
     include_dirs = ['./whisper.cpp/', numpy.get_include()],
     install_requires=[
       'numpy',
-      'ffmpeg-python',
-      'requests'
+      'ffmpeg-python'
     ],
 )

whispercpp.pxd

@@ -10,8 +10,8 @@ cdef nogil:
 	int WHISPER_CHUNK_SIZE = 30
 	int SAMPLE_RATE = 16000
 	char* TEST_FILE = b'test.wav'
-    char* DEFAULT_MODEL = b'ggml-tiny.bin'
-    char* LANGUAGE = b'fr'
+	char* DEFAULT_MODEL = b'ggml-base.bin'
+	char* LANGUAGE = b'en'
 	ctypedef struct audio_data:
 		float* frames;
 		int n_frames;

whispercpp.pyx

@@ -3,43 +3,60 @@
 
 import ffmpeg
 import numpy as np
-import requests
+import urllib.request
 import os
 from pathlib import Path
 
 MODELS_DIR = str(Path('~/.ggml-models').expanduser())
-print("Saving models to:", MODELS_DIR)
 
 
 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* DEFAULT_MODEL = 'base'
+cdef char* LANGUAGE = b'en'
 cdef int N_THREADS = os.cpu_count()
+cdef _Bool PRINT_REALTIME = False
+cdef _Bool PRINT_PROGRESS = False
+cdef _Bool TRANSLATE = False
+
 
 MODELS = {
 	'ggml-tiny.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-tiny.bin',
+	'ggml-tiny.en.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en.bin',
 	'ggml-base.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-base.bin',
+	'ggml-base.en.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin',
 	'ggml-small.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-small.bin',
+	'ggml-small.en.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-small.en.bin',
 	'ggml-medium.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-medium.bin',
+	'ggml-medium.en.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-medium.en.bin',
+	'ggml-large-v1.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-large-v1.bin',
 	'ggml-large.bin': 'https://huggingface.co/datasets/ggerganov/whisper.cpp/resolve/main/ggml-large.bin',
 }
 
-def model_exists(model):
-    return os.path.exists(Path(MODELS_DIR).joinpath(model))
+def model_exists(model, models_dir=MODELS_DIR):
+	return os.path.exists(Path(models_dir).joinpath(model))
 
-def download_model(model):
-    if model_exists(model):
+def download_model(model, models_dir=MODELS_DIR):
+	"""Downloads ggml model with the given identifier
+
+	The filenames mirror the ones given in ggerganov's repos.
+	e.g. 'small' becomes 'ggml-small.bin'
+
+	Args:
+	    model: The model identifier
+	    models_dir: The path where the file is written to
+	"""
+	if model_exists(model, models_dir=models_dir):
 		return
 
-    print(f'Downloading {model}...')
+	print(f'Downloading {model} to {models_dir}...')
 	url = MODELS[model]
-    r = requests.get(url, allow_redirects=True)
-    os.makedirs(MODELS_DIR, exist_ok=True)
-    with open(Path(MODELS_DIR).joinpath(model), 'wb') as f:
-        f.write(r.content)
+	os.makedirs(models_dir, exist_ok=True)
+	with urllib.request.urlopen(url) as r:
+		with open(Path(models_dir).joinpath(model), 'wb') as f:
+			f.write(r.read())
 
 
 cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] load_audio(bytes file, int sr = SAMPLE_RATE):
@@ -57,7 +74,7 @@ cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] load_audio(bytes file, int sr
 				capture_stderr=True
 			)
 		)[0]
-    except:
+	except Exception:
 		raise RuntimeError(f"File '{file}' not found")
 
 	cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] frames = (
@@ -68,43 +85,80 @@ cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] load_audio(bytes file, int sr
 
 	return frames
 
-cdef whisper_full_params default_params() nogil:
+cdef whisper_full_params set_params(_Bool print_realtime, _Bool print_progress, _Bool translate, char* language, int n_threads) nogil:
 	cdef whisper_full_params params = whisper_full_default_params(
 		whisper_sampling_strategy.WHISPER_SAMPLING_GREEDY
 	)
-    params.print_realtime = True
-    params.print_progress = True
-    params.translate = False
-    params.language = <const char *> LANGUAGE
-    n_threads = N_THREADS
+	params.print_realtime = print_realtime
+	params.print_progress = print_progress
+	params.translate = translate
+	params.language = <const char *> language
+	n_threads = n_threads
 	return params
 
-
 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')
-        download_model(model_fullname)
-        model_path = Path(MODELS_DIR).joinpath(model_fullname)
+	def __init__(self, model = DEFAULT_MODEL, models_dir = MODELS_DIR, _Bool print_realtime = PRINT_REALTIME, _Bool print_progress = PRINT_PROGRESS, _Bool translate = TRANSLATE, char* language = LANGUAGE, int n_threads = N_THREADS, _Bool print_system_info = False): # not pretty, look for a way to use kwargs?
+		"""Constructor for Whisper class.
+
+		Automatically checks for model and downloads it if necessary.
+
+		Args:
+		    model: Model identifier, e.g. 'base' (see MODELS)
+		    models_dir: The path where the models should be stored
+		    print_realtime: whisper.cpp's real time transcription output
+		    print_progress: whisper.cpp's progress indicator
+		    translate: whisper.cpp's translation option
+		    language: Which language to use. Must be a byte string.
+		    n_threads: Amount of threads to use
+		    print_system_info: whisper.cpp's system info output
+		"""
+		model_fullname = f'ggml-{model}.bin' #.encode('utf8')
+		download_model(model_fullname, models_dir=models_dir)
+		model_path = Path(models_dir).joinpath(model_fullname)
 		cdef bytes model_b = str(model_path).encode('utf8')
 		self.ctx = whisper_init(model_b)
-        self.params = default_params()
+		self.params = set_params(print_realtime, print_progress, translate, language, n_threads)
+		if print_system_info:
 			whisper_print_system_info()
 
 	def __dealloc__(self):
 		whisper_free(self.ctx)
 
-    def transcribe(self, filename=TEST_FILE):
-        print("Loading data..")
+	def transcribe(self, filename = TEST_FILE):
+		"""Transcribes from given file.
+
+		Args:
+		    filename: Path to file
+
+		Returns:
+		    A result id for extract_text(...)
+
+		Raises:
+		    RuntimeError: The given file could not be found
+		"""
+
+		#print(f"Loading data from '{filename}'...")
 		cdef cnp.ndarray[cnp.float32_t, ndim=1, mode="c"] frames = load_audio(<bytes>filename)
 
-        print("Transcribing..")
+		#print("Transcribing..")
 		return whisper_full(self.ctx, self.params, &frames[0], len(frames))
 	
 	def extract_text(self, int res):
-        print("Extracting text...")
+		"""Extracts the text from a transcription.
+
+		Args:
+		    res: A result id from transcribe(...)
+
+		Results:
+		    A list of transcribed strings.
+
+		Raises:
+		    RuntimeError: The given result id was invalid.
+		"""
+		#print("Extracting text...")
 		if res != 0:
 			raise RuntimeError
 		cdef int n_segments = whisper_full_n_segments(self.ctx)