Skip to content

A lightweight yet powerful audio-to-MIDI converter with pitch bend detection

change minimum_note_length to 127.7ms (11 frames)

Doherty Guirand179939e7e6f8
diff --git a/basic_pitch/inference.py b/basic_pitch/inference.py
index 8591678..3b0032d 100644
--- a/basic_pitch/inference.py
+++ b/basic_pitch/inference.py
@@ -117,6 +117,7 @@ def unwrap_output(output: Tensor, audio_original_length: int, n_overlapping_fram
output_shape = raw_output.shape
n_output_frames_original = int(np.floor(audio_original_length * (ANNOTATIONS_FPS / AUDIO_SAMPLE_RATE)))
unwrapped_output = raw_output.reshape(output_shape[0] * output_shape[1], output_shape[2])
+ print("unwrapped_output:\n",unwrapped_output[:n_output_frames_original, :])
return unwrapped_output[:n_output_frames_original, :] # trim to original audio length
@@ -351,7 +352,7 @@ def predict_and_save(
model_path: Union[pathlib.Path, str] = ICASSP_2022_MODEL_PATH,
onset_threshold: float = 0.5,
frame_threshold: float = 0.3,
- minimum_note_length: float = 58,
+ minimum_note_length: float = 127.70,
minimum_frequency: Optional[float] = None,
maximum_frequency: Optional[float] = None,
multiple_pitch_bends: bool = False,
diff --git a/basic_pitch/note_creation.py b/basic_pitch/note_creation.py
index c8824e9..72a019b 100644
--- a/basic_pitch/note_creation.py
+++ b/basic_pitch/note_creation.py
@@ -48,7 +48,7 @@ def model_output_to_notes(
onset_thresh: float,
frame_thresh: float,
infer_onsets: bool = True,
- min_note_len: int = 5,
+ min_note_len: int = 11,
min_freq: Optional[float] = None,
max_freq: Optional[float] = None,
include_pitch_bends: bool = True,
@@ -396,6 +396,7 @@ def output_to_notes_polyphonic(
# loop over onsets
note_events = []
for note_start_idx, freq_idx in zip(onset_time_idx, onset_freq_idx):
+ print("freq_idx: ", freq_idx)
# if we're too close to the end of the audio, continue
if note_start_idx >= n_frames - 1:
continue
@@ -494,5 +495,6 @@ def output_to_notes_polyphonic(
amplitude,
)
)
-
+ print(note_events)
+
return note_events