Set MP3 encoder preset (facebookresearch#534)

Author: CarlGao4

README.md

@@ -184,6 +184,7 @@ In order to try Demucs, you can just run from any folder (as long as you properl
 demucs PATH_TO_AUDIO_FILE_1 [PATH_TO_AUDIO_FILE_2 ...]   # for Demucs
 # If you used `pip install --user` you might need to replace demucs with python3 -m demucs
 python3 -m demucs --mp3 --mp3-bitrate BITRATE PATH_TO_AUDIO_FILE_1  # output files saved as MP3
+        # use --mp3-preset to change encoder preset, 2 for best quality, 7 for fastest
 # If your filename contain spaces don't forget to quote it !!!
 demucs "my music/my favorite track.mp3"
 # You can select different models with `-n` mdx_q is the quantized model, smaller but maybe a bit less accurate.

demucs/__init__.py

@@ -4,4 +4,4 @@
 # This source code is licensed under the license found in the
 # LICENSE file in the root directory of this source tree.
 
-__version__ = "4.0.1a2"
+__version__ = "4.0.1a3"

demucs/audio.py

@@ -196,15 +196,15 @@ def as_dtype_pcm(wav, dtype):
         return i16_pcm(wav)
 
 
-def encode_mp3(wav, path, samplerate=44100, bitrate=320, verbose=False):
+def encode_mp3(wav, path, samplerate=44100, bitrate=320, quality=2, verbose=False):
     """Save given audio as mp3. This should work on all OSes."""
     C, T = wav.shape
     wav = i16_pcm(wav)
     encoder = lameenc.Encoder()
     encoder.set_bit_rate(bitrate)
     encoder.set_in_sample_rate(samplerate)
     encoder.set_channels(C)
-    encoder.set_quality(2)  # 2-highest, 7-fastest
+    encoder.set_quality(quality)  # 2-highest, 7-fastest
     if not verbose:
         encoder.silence()
     wav = wav.data.cpu()
@@ -239,16 +239,18 @@ def save_audio(wav: torch.Tensor,
                bitrate: int = 320,
                clip: tp.Literal["rescale", "clamp", "tanh", "none"] = 'rescale',
                bits_per_sample: tp.Literal[16, 24, 32] = 16,
-               as_float: bool = False):
+               as_float: bool = False,
+               preset: tp.Literal[2, 3, 4, 5, 6, 7] = 2):
     """Save audio file, automatically preventing clipping if necessary
     based on the given `clip` strategy. If the path ends in `.mp3`, this
-    will save as mp3 with the given `bitrate`.
+    will save as mp3 with the given `bitrate`. Use `preset` to set mp3 quality:
+    2 for highest quality, 7 for fastest speed
     """
     wav = prevent_clip(wav, mode=clip)
     path = Path(path)
     suffix = path.suffix.lower()
     if suffix == ".mp3":
-        encode_mp3(wav, path, samplerate, bitrate, verbose=True)
+        encode_mp3(wav, path, samplerate, bitrate, preset, verbose=True)
     elif suffix == ".wav":
         if as_float:
             bits_per_sample = 32

demucs/separate.py

@@ -113,6 +113,9 @@ def get_parser():
                         default=320,
                         type=int,
                         help="Bitrate of converted mp3.")
+    parser.add_argument("--mp3-preset", choices=range(2, 8), type=int, default=2,
+                        help="Encoder preset of MP3, 2 for highest quality, 7 for "
+                        "fastest speed. Default is 2")
     parser.add_argument("-j", "--jobs",
                         default=0,
                         type=int,
@@ -182,6 +185,7 @@ def main(opts=None):
         kwargs = {
             'samplerate': model.samplerate,
             'bitrate': args.mp3_bitrate,
+            'preset': args.mp3_preset,
             'clip': args.clip_mode,
             'as_float': args.float32,
             'bits_per_sample': 24 if args.int24 else 16,

docs/release.md

@@ -15,6 +15,8 @@ Will use CPU for complex numbers, when using MPS device (all other computations
 
 Optimize codes to save memory
 
+Allow changing preset of MP3
+
 ## V4.0.0, 7th of December 2022
 
 Adding hybrid transformer Demucs model.