| name | wandas-signal-processing |
| description | Use when applying filters (lowpass, highpass, bandpass, A-weighting), normalizing signals, resampling, trimming, adding fades, computing RMS trends, calculating sound level (dB, A-weighting), or computing psychoacoustic metrics (loudness, roughness, sharpness) with wandas. |
wandas: Signal Processing
ChannelFrame の時間域処理。各メソッドは新しいフレームを返す（イミュータブル）。
Mandatory Rules
- Wandas-first:
scipy.signal.butter 等の直接呼び出し禁止。必ず wandas メソッドを使う。
- Method chaining: 各メソッドは新しいフレームを返す。元のフレームは変更されない。
- Visualization:
.plot() / .describe() を使う。plt.plot(frame.data) は禁止。
フィルタ・前処理
| メソッド | シグネチャ | 返り値 |
|---|
.high_pass_filter | (cutoff: float, order: int = 4) | ChannelFrame |
.low_pass_filter | (cutoff: float, order: int = 4) | ChannelFrame |
.band_pass_filter | (low_cutoff: float, high_cutoff: float, order: int = 4) | ChannelFrame |
.a_weighting | () | ChannelFrame |
.normalize | (norm: float = inf, axis: int = -1, threshold=None, fill=None) | ChannelFrame |
.remove_dc | () | ChannelFrame |
.resampling | (target_sr: float) | ChannelFrame |
.trim | (start: float = 0, end: float = None) | ChannelFrame |
.fix_length | (length: int = None, duration: float = None) | ChannelFrame |
.fade | (fade_ms: float = 50) | ChannelFrame |
.hpss_harmonic | (kernel_size: int = 31) | ChannelFrame |
.hpss_percussive | (kernel_size: int = 31) | ChannelFrame |
音圧レベル・トレンド
| メソッド | シグネチャ | 返り値 | 注意 |
|---|
.sound_level | (freq_weighting: str = "Z", time_weighting: str = "Fast", dB: bool = False) | ChannelFrame | dB=True で dB 換算。ch_units=['Pa'] 設定時のみ正しい dB SPL |
.rms_trend | (frame_length: int = 2048, hop_length: int = 512, dB: bool = False, Aw: bool = False) | ChannelFrame | フレームごとの時変 RMS |
freq_weighting オプション: "A", "C", "Z"（フラット）
time_weighting オプション: "Fast"（125ms）, "Slow"（1s）
心理音響指標
| メソッド | シグネチャ | 返り値 | 単位 |
|---|
.loudness_zwtv | (field_type: str = "free") | ChannelFrame | sone（ISO 532-1） |
.loudness_zwst | (field_type: str = "free") | NDArrayReal | sone（スカラー値） |
.roughness_dw | (overlap: float = 0.5) | ChannelFrame | asper |
.roughness_dw_spec | (overlap: float = 0.5) | RoughnessFrame | asper/Bark |
.sharpness_din | (weighting: str = "din", field_type: str = "free") | ChannelFrame | acum |
.sharpness_din_st | (weighting: str = "din", field_type: str = "free") | NDArrayReal | acum（スカラー値） |
⚠️ loudness_zwst と sharpness_din_st は フレームではなく NDArrayReal を返す。.plot() はチェーンできない。
field_type オプション: "free"（自由音場）, "diffuse"（拡散音場）
weighting オプション: "din", "aures", "bismarck", "fastl"
Patterns
ノイズ除去パイプライン
import wandas as wd
signal = wd.read_wav("noisy.wav")
cleaned = (signal
.high_pass_filter(cutoff=50)
.low_pass_filter(cutoff=8000)
.normalize()
.fade(fade_ms=10))
cleaned.describe()
音圧レベル dB(A) の計算
import wandas as wd
import numpy as np
frame = wd.from_numpy(signal_pa, sampling_rate=sr, ch_units=["Pa"])
spl = frame.sound_level(freq_weighting="A", time_weighting="Fast", dB=True)
print(f"Leq: {np.mean(spl.data):.1f} dB(A)")
print(f"Lmax: {np.max(spl.data):.1f} dB(A)")
心理音響総合評価
import wandas as wd
sig = wd.read_wav("audio.wav")
loudness = sig.loudness_zwtv(field_type="free")
roughness = sig.roughness_dw(overlap=0.5)
sharpness = sig.sharpness_din(weighting="din")
loudness.plot(title="Time-varying Loudness [sone]")
roughness.plot(title="Time-varying Roughness [asper]")
sharpness.plot(title="Time-varying Sharpness [acum]")
loudness_val = sig.loudness_zwst(field_type="free")
sharpness_val = sig.sharpness_din_st(weighting="din")
print(f"Loudness: {loudness_val} sone")
print(f"Sharpness: {sharpness_val} acum")
RMS トレンド（A 重み付き）
rms = frame.rms_trend(frame_length=2048, hop_length=512, dB=True, Aw=True)
rms.plot(title="A-weighted RMS Trend")
HPSS による楽音・打楽音分離
import wandas as wd
signal = wd.read_wav("music.wav")
harmonic = signal.hpss_harmonic()
percussive = signal.hpss_percussive()
harmonic.describe()
Common Mistakes
| 間違い | 正解 |
|---|
.band_pass_filter(low=100, high=5000) | .band_pass_filter(low_cutoff=100, high_cutoff=5000) |
.fft().a_weighting() | a_weighting() は ChannelFrame にのみ適用可能 |
normalize() が RMS 正規化 | デフォルトはピーク正規化（norm=inf） |
| カットオフをナイキスト以上に設定 | cutoff < sampling_rate / 2 を確認 |
loudness_zwst().plot() を呼ぶ | loudness_zwst は NDArrayReal を返す。フレームではない |
sound_level(dB=True) の値がおかしい | ch_units=['Pa'] 未設定時は参照値 1.0 で計算される |
Documentation Map
