Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    RuntimeError
Message:      Failed to open input buffer: Invalid data found when processing input
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                         ^^^^^^^^^
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                         ^^^^^^^^^^^^^^^^^^^^^
                File "/src/services/worker/src/worker/utils.py", line 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2567, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2103, in __iter__
                  batch = formatter.format_batch(pa_table)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/formatting/formatting.py", line 472, in format_batch
                  batch = self.python_features_decoder.decode_batch(batch)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/formatting/formatting.py", line 234, in decode_batch
                  return self.features.decode_batch(batch, token_per_repo_id=self.token_per_repo_id) if self.features else batch
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/features/features.py", line 2254, in decode_batch
                  decode_nested_example(self[column_name], value, token_per_repo_id=token_per_repo_id)
                File "/usr/local/lib/python3.12/site-packages/datasets/features/features.py", line 1508, in decode_nested_example
                  return schema.decode_example(obj, token_per_repo_id=token_per_repo_id) if obj is not None else None
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/features/audio.py", line 211, in decode_example
                  audio = AudioDecoder(
                          ^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/torchcodec/decoders/_audio_decoder.py", line 64, in __init__
                  self._decoder = create_decoder(source=source, seek_mode="approximate")
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/torchcodec/decoders/_decoder_utils.py", line 45, in create_decoder
                  return core.create_from_file_like(source, seek_mode)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/torchcodec/_core/ops.py", line 151, in create_from_file_like
                  return _convert_to_tensor(_pybind_ops.create_from_file_like(file_like, seek_mode))
                                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
              RuntimeError: Failed to open input buffer: Invalid data found when processing input

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

SFZ Instruments

Select open source musical instruments which use the SFZ format


Attribution

All SFZ instruments were sourced from sfzinstruments


How to use (Linux/Ubuntu)

1) Unzip the dataset into some folder

2) Setup environment

!sudo apt install libsndfile1 ffmpeg libjack-jackd2-dev -y
!pip install soundfile numpy scipy tqdm

3) Build and install sfizz

!sudo apt install cmake g++ git libsndfile1-dev libjack-jackd2-dev \
                 libsamplerate0-dev libboost-dev libzstd-dev \
                 libcurl4-openssl-dev libx11-dev -y

!git clone https://github.com/sfztools/sfizz.git
%cd sfizz
!mkdir build
%cd build
!cmake .. -DCMAKE_BUILD_TYPE=Release
!make -j$(nproc)
!sudo make install
!sudo ldconfig

4) Use the following python script to render your MIDIs

#!/usr/bin/env python3
"""
render_and_mix.py

Render MIDI+SFZ pairs using sfizz_render, mix them, apply basic mastering (limiter + normalize),
and optionally run ffmpeg loudness normalization.

Dependencies:
  - sfizz_render (system binary)
  - libsndfile (system)
  - Python packages: soundfile, numpy, scipy, tqdm
  - Optional: ffmpeg (for LUFS normalization)
"""

import os
import shutil
import subprocess
import tempfile
from concurrent.futures import ThreadPoolExecutor, as_completed
from pathlib import Path
from typing import Dict, Optional

import numpy as np
import soundfile as sf
from scipy.signal import fftconvolve
from tqdm import tqdm

# ---------- Utility audio functions ----------

def db_to_linear(db: float) -> float:
    return 10.0 ** (db / 20.0)

def linear_to_db(x: float) -> float:
    return 20.0 * np.log10(np.maximum(x, 1e-12))

def apply_pan(stereo: np.ndarray, pan: float) -> np.ndarray:
    # pan: -1 (left) .. +1 (right)
    left_gain = np.cos((pan + 1) * (np.pi / 4))
    right_gain = np.sin((pan + 1) * (np.pi / 4))
    stereo[:, 0] *= left_gain
    stereo[:, 1] *= right_gain
    return stereo

def ensure_stereo(arr: np.ndarray) -> np.ndarray:
    if arr.ndim == 1:
        return np.stack([arr, arr], axis=1)
    if arr.shape[1] == 1:
        return np.repeat(arr, 2, axis=1)
    return arr[:, :2]

def soft_limiter(signal: np.ndarray, threshold: float = 0.98, release: float = 0.01, sample_rate: int = 48000) -> np.ndarray:
    # Simple per-sample soft clipping with smoothing
    out = np.copy(signal)
    # apply tanh-style soft clip scaled to threshold
    scale = 1.0 / threshold
    out = np.tanh(out * scale) / scale
    return out

def normalize_peak(signal: np.ndarray, target_dbfs: float = -1.0) -> np.ndarray:
    peak = np.max(np.abs(signal))
    if peak <= 0:
        return signal
    target_lin = db_to_linear(target_dbfs)
    gain = target_lin / peak
    return signal * gain

# ---------- sfizz_render wrapper ----------

def find_sfizz_render() -> Optional[str]:
    # Try common binary names
    for name in ("sfizz_render", "sfizz-render", "sfizz_render.exe"):
        path = shutil.which(name)
        if path:
            return path
    return None

def render_with_sfizz(sfizz_bin: str, midi_path: str, sfz_path: str, out_wav: str,
                      sample_rate: int = 48000, quality: int = 3, polyphony: int = 256,
                      use_eot: bool = True, verbose: bool = False) -> None:
    cmd = [
        sfizz_bin,
        "--midi", str(midi_path),
        "--sfz", str(sfz_path),
        "--wav", str(out_wav),
        "--samplerate", str(sample_rate),
        "--quality", str(quality),
        "--polyphony", str(polyphony),
    ]
    if use_eot:
        cmd.append("--use-eot")
    if verbose:
        cmd.append("--verbose")
    # Run and raise on error
    subprocess.run(cmd, check=True)

# ---------- Main render and mix function ----------

def render_and_mix(
    midi_sfz_map: Dict[str, str],
    out_path: str,
    *,
    sample_rate: int = 48000,
    quality: int = 3,
    polyphony: int = 256,
    track_options: Optional[Dict[str, Dict]] = None,
    normalize_lufs: Optional[float] = None,
    use_eot: bool = True,
    workers: int = 2,
    verbose: bool = False
) -> None:
    """
    Render each MIDI->SFZ pair, mix, post-process, and write final WAV to out_path.
    """
    sfizz_bin = find_sfizz_render()
    if not sfizz_bin:
        raise FileNotFoundError("sfizz_render binary not found in PATH. Install sfizz-render first.")

    tmpdir = Path(tempfile.mkdtemp(prefix="sfizz_render_"))
    rendered_files = {}

    # Render in parallel
    with ThreadPoolExecutor(max_workers=workers) as ex:
        futures = {}
        for midi, sfz in midi_sfz_map.items():
            midi_p = Path(midi)
            sfz_p = Path(sfz)
            if not midi_p.exists():
                raise FileNotFoundError(f"MIDI file not found: {midi}")
            if not sfz_p.exists():
                raise FileNotFoundError(f"SFZ file not found: {sfz}")
            out_wav = tmpdir / (midi_p.stem + "_" + sfz_p.stem + ".wav")
            futures[ex.submit(render_with_sfizz, sfizz_bin, str(midi_p), str(sfz_p), str(out_wav),
                              sample_rate, quality, polyphony, use_eot, verbose)] = (midi, str(out_wav))

        # Wait and collect
        for fut in tqdm(as_completed(futures), total=len(futures), desc="Rendering"):
            midi_key, wav_path = futures[fut]
            fut.result()  # will raise if render failed
            rendered_files[midi_key] = wav_path

    # Load and align
    tracks = []
    max_len = 0
    for midi_key, wav_path in rendered_files.items():
        data, sr = sf.read(wav_path, always_2d=True)
        if sr != sample_rate:
            # resample if needed (simple linear resample)
            import math
            ratio = sample_rate / sr
            new_len = int(math.ceil(data.shape[0] * ratio))
            # use scipy.signal.resample for decent quality
            from scipy.signal import resample
            data = resample(data, new_len, axis=0)
        data = ensure_stereo(data)
        tracks.append((midi_key, data))
        if data.shape[0] > max_len:
            max_len = data.shape[0]

    # Prepare final mix buffer
    mix = np.zeros((max_len, 2), dtype=np.float32)

    # Apply per-track options and mix
    for midi_key, data in tracks:
        opts = (track_options or {}).get(midi_key, {})
        gain_db = float(opts.get("gain_db", 0.0))
        pan = float(opts.get("pan", 0.0))
        gain_lin = db_to_linear(gain_db)
        # pad to max_len
        pad_len = max_len - data.shape[0]
        if pad_len > 0:
            data = np.vstack([data, np.zeros((pad_len, 2), dtype=data.dtype)])
        data = data.astype(np.float32) * gain_lin
        data = apply_pan(data, pan)
        mix[:data.shape[0], :] += data

    # Basic safety: prevent NaNs/Infs
    mix = np.nan_to_num(mix, nan=0.0, posinf=0.0, neginf=0.0)

    # Apply soft limiter and normalization
    mix = soft_limiter(mix, threshold=0.98, sample_rate=sample_rate)
    mix = normalize_peak(mix, target_dbfs=-1.0)

    # Write intermediate file
    intermediate = tmpdir / "mixed_intermediate.wav"
    sf.write(str(intermediate), mix, samplerate=sample_rate, subtype="PCM_24")

    # Optional LUFS normalization via ffmpeg loudnorm
    final_out = Path(out_path)
    if normalize_lufs is not None:
        ffmpeg = shutil.which("ffmpeg")
        if not ffmpeg:
            raise FileNotFoundError("ffmpeg not found but normalize_lufs requested.")
        # two-pass loudnorm recommended; here we do a single-pass approximate target
        cmd = [
            ffmpeg, "-y", "-i", str(intermediate),
            "-af", f"loudnorm=I={normalize_lufs}:TP=-1.5:LRA=11",
            "-ar", str(sample_rate),
            "-ac", "2",
            "-c:a", "pcm_s24le",
            str(final_out)
        ]
        subprocess.run(cmd, check=True)
    else:
        # move intermediate to final
        shutil.move(str(intermediate), str(final_out))

    # cleanup
    try:
        shutil.rmtree(tmpdir)
    except Exception:
        pass

# ---------- Example usage ----------
if __name__ == "__main__":
    # Example mapping: two MIDI files each with their SFZ instrument
    mapping = {
        "midi/drums.mid": "sfz/drumkit.sfz",
        "midi/piano.mid": "sfz/grand_piano.sfz",
    }
    track_opts = {
        "midi/drums.mid": {"gain_db": -1.5, "pan": 0.0},
        "midi/piano.mid": {"gain_db": -3.0, "pan": -0.1},
    }
    render_and_mix(mapping, "final_mix.wav", sample_rate=48000, quality=3, polyphony=256,
                   track_options=track_opts, normalize_lufs=-14.0, use_eot=True, workers=2, verbose=False)

Project Los Angeles

Tegridy Code 2026

Downloads last month
8