Pierfrancesco/adt-str

ADT_STR

Automatic Drum Transcription model exported from this repository.

Default variant: setting-tau-0.8.

Quick start

from huggingface_hub import snapshot_download
import sys

repo_dir = snapshot_download("Pierfrancesco/adt-str")
sys.path.insert(0, repo_dir)

from adt_transcriber import ADTTranscriber

transcriber = ADTTranscriber.from_pretrained(repo_dir, variant="setting-tau-0.8")
midi_path = transcriber.transcribe("path/to/audio.wav", output_dir="outputs")
transcriber.play(midi_path, output_path="outputs/preview.wav")

transcribe also accepts a list of audio paths:

midi_paths = transcriber.transcribe(
    ["path/to/audio_1.wav", "path/to/audio_2.wav"],
    output_dir="outputs",
)

or a padded tensor batch:

midi_paths = transcriber.transcribe(
    padded_audio_batch,
    sample_rate=24000,
    lengths=valid_lengths,
    output_dir="outputs",
)

Variants

Variant	Folder	Parameters
setting-tau-0.4	setting-tau-0.4	69,000,824
setting-tau-0.6	setting-tau-0.6	69,000,824
setting-tau-0.8	setting-tau-0.8	69,000,824

Local loading

from adt_transcriber import ADTTranscriber

transcriber = ADTTranscriber.from_pretrained(".", variant="setting-tau-0.8")

You can choose setting-tau-0.4, setting-tau-0.6, or setting-tau-0.8 with the variant argument.