Merge pull request #300 from rakuri255/fix-bpm-with-wav-and-whisper-force-cpu-and-upgrade-librosa

fix BPM, whisper force cpu and librosa upgrade
This commit is contained in:
Vadim Rangnau authored and GitHub committed 2026-02-16 00:29:36 +01:00
commit e07cb90242
6 files changed
+22 -14

No files matched your search

+2 -3
View File
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "UltraSinger"
version = "0.0.13.dev14"
version = "0.0.13.dev15"
description = "A tool to create UltraStar karaoke files from audio files"
requires-python = ">=3.12"
dependencies = [
@@ -16,8 +16,7 @@ dependencies = [
"langcodes",
"language_data",
"packaging",
"librosa>=0.10.2",
"numba>=0.59.0",
"librosa",
"swift-f0",
"pydub",
"demucs",
+1 -1
View File
@@ -11,7 +11,7 @@ from modules.Ultrastar.ultrastar_txt import FormatVersion
@dataclass
class Settings:
APP_VERSION = "0.0.13.dev14"
APP_VERSION = "0.0.13.dev15"
CONFIDENCE_THRESHOLD = 0.6
CONFIDENCE_PROMPT_TIMEOUT = 4
+9 -5
View File
@@ -171,6 +171,10 @@ def run() -> tuple[str, Score, Score]:
# Create process audio
process_data.process_data_paths.processing_audio_path = CreateProcessAudio(process_data)
# Get BPM from wav file
if not settings.input_file_is_ultrastar_txt:
process_data.media_info.bpm = get_bpm_from_file(process_data.process_data_paths.processing_audio_path)
# Detect key
detected_key, detected_mode = detect_key_from_audio(process_data.process_data_paths.processing_audio_path)
if process_data.media_info.music_key is None:
@@ -611,15 +615,17 @@ def transcribe_audio(cache_folder_path: str, processing_audio_path: str) -> Tran
transcription_result = None
whisper_align_model_string = None
if settings.transcriber == "whisper":
if not settings.whisper_align_model is None: whisper_align_model_string = settings.whisper_align_model.replace("/", "_")
transcription_config = f"{settings.transcriber}_{settings.whisper_model.value}_{settings.pytorch_device}_{whisper_align_model_string}_{settings.whisper_batch_size}_{settings.whisper_compute_type}_{settings.language}"
if not settings.whisper_align_model is None:
whisper_align_model_string = settings.whisper_align_model.replace("/", "_")
whisper_device = "cpu" if settings.force_whisper_cpu else settings.pytorch_device
transcription_config = f"{settings.transcriber}_{settings.whisper_model.value}_{whisper_device}_{whisper_align_model_string}_{settings.whisper_batch_size}_{settings.whisper_compute_type}_{settings.language}"
transcription_path = os.path.join(cache_folder_path, f"{transcription_config}.json")
cached_transcription_available = check_file_exists(transcription_path)
if settings.skip_cache_transcription or not cached_transcription_available:
transcription_result = transcribe_with_whisper(
processing_audio_path,
settings.whisper_model,
settings.pytorch_device,
whisper_device,
settings.whisper_align_model,
settings.whisper_batch_size,
settings.whisper_compute_type,
@@ -684,7 +690,6 @@ def infos_from_audio_video_input_file() -> tuple[str, str, str, MediaInfo]:
if song_info.cover_image_data is not None:
save_image(song_info.cover_image_data, basename_without_ext, song_folder_output_path)
real_bpm = get_bpm_from_file(ultrastar_audio_input_path)
return (
basename_without_ext,
song_folder_output_path,
@@ -694,7 +699,6 @@ def infos_from_audio_video_input_file() -> tuple[str, str, str, MediaInfo]:
title=song_info.title,
year=song_info.year,
genre=song_info.genres,
bpm=real_bpm,
cover_url=song_info.cover_url,
audio_extension=audio_ext,
video_extension=video_ext
+9 -2
View File
@@ -1,4 +1,5 @@
import librosa
import soundfile as sf
from modules.console_colors import ULTRASINGER_HEAD, blue_highlighted
@@ -6,7 +7,7 @@ from modules.console_colors import ULTRASINGER_HEAD, blue_highlighted
def get_bpm_from_data(data, sampling_rate):
"""Get real bpm from audio data"""
onset_env = librosa.onset.onset_strength(y=data, sr=sampling_rate)
wav_tempo = librosa.beat.tempo(onset_envelope=onset_env, sr=sampling_rate)
wav_tempo = librosa.feature.tempo(onset_envelope=onset_env, sr=sampling_rate)
print(f"{ULTRASINGER_HEAD} BPM is {blue_highlighted(str(round(wav_tempo[0], 2)))}")
return wav_tempo[0]
@@ -14,5 +15,11 @@ def get_bpm_from_data(data, sampling_rate):
def get_bpm_from_file(wav_file: str) -> float:
"""Get real bpm from audio file"""
data, sampling_rate = librosa.load(wav_file, sr=None)
data, sampling_rate = sf.read(wav_file, dtype='float32')
# Transpose if stereo to match librosa's expected format
if len(data.shape) > 1:
data = data.T
# Convert to mono if stereo
if data.ndim > 1:
data = librosa.to_mono(data)
return get_bpm_from_data(data, sampling_rate)
-2
View File
@@ -116,7 +116,6 @@ def download_from_youtube(input_url: str, output_folder_path: str, cookiefile: s
input_url, basename_without_ext, song_output, cookiefile
)
real_bpm = get_bpm_from_file(audio_file_path)
return (
basename_without_ext,
song_output,
@@ -126,7 +125,6 @@ def download_from_youtube(input_url: str, output_folder_path: str, cookiefile: s
title=song_info.title,
year=song_info.year,
genre=song_info.genres,
bpm=real_bpm,
cover_url=cover_url,
video_url=input_url,
audio_extension=audio_ext,
+1 -1
View File
@@ -19,7 +19,7 @@ class MediaInfo:
"""Media Info"""
title: str
artist: str
bpm: float
bpm: Optional[float] = None
year: Optional[str] = None
genre: Optional[str] = None
language: Optional[str] = None