#903: downmix 5.1 and other multichannel audio to stereo
Problem: footage audio kept only its first two channels, so the centre, LFE and surrounds of a 5.1 file were dropped. AAC 5.1 was worse: FilmCraft's AAC-LC decoder delivers AAC's own order (C L R Ls Rs LFE), so the left output was the centre and front-right was lost. Fix: crates/media/src/downmix.rs holds one ITU-R BS.775 stereo downmix (centre and surrounds at -3 dB, LFE left out, fixed normalisation by the largest row sum so full scale cannot clip), with channel orders for 3 to 8 channels in WAVE/SMPTE order and AAC's native order, and only the front pair for more than 8. MediaPool::read_stereo applies it, which preview, waveform and export mixdown all read through. Mono and stereo are unchanged. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
vangel76
parent
67434e9038
commit
64fabbb2fc
|
|
@ -0,0 +1,133 @@
|
|||
//! Stereo downmix of multichannel audio, shared by preview, waveform and export mixdown (they
|
||||
//! all read footage audio through `MediaPool::audio_samples`).
|
||||
//!
|
||||
//! The matrix is the usual ITU-R BS.775 one: front L/R at unity, the centre and every surround
|
||||
//! or back channel at −3 dB (1/√2) into both sides' own side (centre into both), the low
|
||||
//! frequency effects channel left out (as ffmpeg's and swresample's default, `lfe_mix_level` 0;
|
||||
//! not verified against Adobe's documentation, which does not state it). A back-centre channel
|
||||
//! reaches both sides at −6 dB. The whole matrix is then scaled by one fixed gain, the reciprocal
|
||||
//! of the largest row sum (ffmpeg's default normalisation), so that full-scale input can never
|
||||
//! exceed full scale: 5.1 is scaled by 1/(1 + 0.707 + 0.707). Mono stays duplicated at unity and
|
||||
//! stereo is untouched.
|
||||
//!
|
||||
//! Channel order follows the WAVE / SMPTE convention by channel count (what FilmCraft's AC-3, PCM,
|
||||
//! FLAC and Opus readers deliver): 3 = L R C, 4 = quad L R Ls Rs, 5 = L R C Ls Rs,
|
||||
//! 6 = L R C LFE Ls Rs, 7 = L R C LFE Cs Ls Rs, 8 = L R C LFE Lb Rb Ls Rs. FilmCraft's AAC-LC
|
||||
//! decoder instead delivers AAC's own order (C L R first, LFE last), so AAC gets its own tables.
|
||||
//! More than 8 channels carry no known layout: only the first two (front L/R) are used.
|
||||
|
||||
use std::f32::consts::FRAC_1_SQRT_2;
|
||||
|
||||
/// A source channel's contribution to the left and right outputs, before normalisation.
|
||||
type Pair = [f32; 2];
|
||||
|
||||
const FL: Pair = [1.0, 0.0];
|
||||
const FR: Pair = [0.0, 1.0];
|
||||
const FC: Pair = [FRAC_1_SQRT_2, FRAC_1_SQRT_2];
|
||||
const LFE: Pair = [0.0, 0.0];
|
||||
const SL: Pair = [FRAC_1_SQRT_2, 0.0];
|
||||
const SR: Pair = [0.0, FRAC_1_SQRT_2];
|
||||
const CS: Pair = [0.5, 0.5];
|
||||
|
||||
/// The per-channel `[left, right]` gains for `count` channels in the given order, normalised.
|
||||
/// `aac` selects AAC's native channel order. Empty for zero channels.
|
||||
pub fn stereo_gains(count: usize, aac: bool) -> Vec<Pair> {
|
||||
let raw: &[Pair] = match (count, aac) {
|
||||
(0, _) => &[],
|
||||
(1, _) => &[[1.0, 1.0]],
|
||||
(2, _) => &[FL, FR],
|
||||
(3, false) => &[FL, FR, FC],
|
||||
(3, true) => &[FC, FL, FR],
|
||||
(4, false) => &[FL, FR, SL, SR],
|
||||
(4, true) => &[FC, FL, FR, CS],
|
||||
(5, false) => &[FL, FR, FC, SL, SR],
|
||||
(5, true) => &[FC, FL, FR, SL, SR],
|
||||
(6, false) => &[FL, FR, FC, LFE, SL, SR],
|
||||
(6, true) => &[FC, FL, FR, SL, SR, LFE],
|
||||
(7, false) => &[FL, FR, FC, LFE, CS, SL, SR],
|
||||
(7, true) => &[FC, FL, FR, SL, SR, CS, LFE],
|
||||
(8, false) => &[FL, FR, FC, LFE, SL, SR, SL, SR],
|
||||
(8, true) => &[FC, FL, FR, SL, SR, SL, SR, LFE],
|
||||
_ => &[FL, FR],
|
||||
};
|
||||
let mut out = raw.to_vec();
|
||||
// Channels beyond a known layout (more than 8) are silent.
|
||||
out.resize(count, [0.0, 0.0]);
|
||||
let peak = out.iter().map(|g| g[0]).sum::<f32>().max(out.iter().map(|g| g[1]).sum::<f32>());
|
||||
// Mono is deliberately left at unity on both sides; so is anything already within range.
|
||||
if count > 2 && peak > 1.0 {
|
||||
for g in &mut out {
|
||||
g[0] /= peak;
|
||||
g[1] /= peak;
|
||||
}
|
||||
}
|
||||
out
|
||||
}
|
||||
|
||||
/// Downmix planar `channels` to interleaved stereo in `out` (`out.len() / 2` frames); frames a
|
||||
/// channel is too short for count as silence, extra frames in `out` stay as they are.
|
||||
pub fn downmix_into(channels: &[Vec<f32>], aac: bool, out: &mut [f32]) {
|
||||
let gains = stereo_gains(channels.len(), aac);
|
||||
for (ch, g) in channels.iter().zip(&gains) {
|
||||
if g[0] == 0.0 && g[1] == 0.0 {
|
||||
continue;
|
||||
}
|
||||
for (o, v) in out.as_chunks_mut::<2>().0.iter_mut().zip(ch) {
|
||||
o[0] += v * g[0];
|
||||
o[1] += v * g[1];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
|
||||
#[test]
|
||||
fn every_channel_count_has_a_bounded_defined_matrix() {
|
||||
for aac in [false, true] {
|
||||
for n in 0..40usize {
|
||||
let g = stereo_gains(n, aac);
|
||||
assert_eq!(g.len(), n);
|
||||
let (l, r) = (g.iter().map(|p| p[0]).sum::<f32>(), g.iter().map(|p| p[1]).sum::<f32>());
|
||||
if n > 2 {
|
||||
assert!(l <= 1.0001 && r <= 1.0001, "{n} channels (aac {aac}): {l} {r}");
|
||||
assert!(l > 0.0 && r > 0.0);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn five_one_matches_the_bs775_matrix() {
|
||||
let k = 1.0 / (1.0 + 2.0 * FRAC_1_SQRT_2);
|
||||
let g = stereo_gains(6, false);
|
||||
assert!((g[0][0] - k).abs() < 1e-6 && g[0][1] == 0.0);
|
||||
assert!((g[2][0] - k * FRAC_1_SQRT_2).abs() < 1e-6 && (g[2][1] - k * FRAC_1_SQRT_2).abs() < 1e-6);
|
||||
assert_eq!(g[3], [0.0, 0.0], "LFE is left out");
|
||||
assert!((g[4][0] - k * FRAC_1_SQRT_2).abs() < 1e-6 && g[4][1] == 0.0);
|
||||
assert!((g[5][1] - k * FRAC_1_SQRT_2).abs() < 1e-6 && g[5][0] == 0.0);
|
||||
// AAC order: the centre is first, LFE last
|
||||
let a = stereo_gains(6, true);
|
||||
assert_eq!((a[0], a[5]), (g[2], g[3]));
|
||||
assert_eq!((a[1], a[2]), (g[0], g[1]));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn mono_and_stereo_are_untouched_and_wide_layouts_use_the_front_pair() {
|
||||
assert_eq!(stereo_gains(1, false), vec![[1.0, 1.0]]);
|
||||
assert_eq!(stereo_gains(2, true), vec![FL, FR]);
|
||||
let g = stereo_gains(12, false);
|
||||
assert_eq!((g[0], g[1], g[2]), (FL, FR, [0.0, 0.0]));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn ragged_and_empty_input_never_panics() {
|
||||
let mut out = vec![0.0; 8];
|
||||
downmix_into(&[], false, &mut out);
|
||||
downmix_into(&[vec![1.0; 100], vec![1.0; 2], vec![], vec![1.0; 3], vec![1.0; 1], vec![1.0]], true, &mut out);
|
||||
assert!(out.iter().all(|v| v.is_finite()));
|
||||
let mut none: [f32; 0] = [];
|
||||
downmix_into(&vec![vec![1.0; 3]; 7], false, &mut none);
|
||||
}
|
||||
}
|
||||
|
|
@ -16,6 +16,7 @@
|
|||
#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)]
|
||||
|
||||
mod convert;
|
||||
pub mod downmix;
|
||||
pub mod exr_channels;
|
||||
mod gif_anim;
|
||||
mod layered;
|
||||
|
|
|
|||
|
|
@ -448,8 +448,8 @@ impl MediaPool {
|
|||
}
|
||||
|
||||
/// `frames` stereo sample frames of `footage`'s audio starting at source time `start`, at
|
||||
/// `rate` Hz, interleaved (L R L R …). Mono is duplicated to both channels; channels beyond
|
||||
/// the first two are dropped. Silence where there is no audio (or before the media starts).
|
||||
/// `rate` Hz, interleaved (L R L R …). Mono is duplicated to both channels; multichannel
|
||||
/// sources are downmixed to stereo ([`crate::downmix`]). Silence where there is no audio (or before the media starts).
|
||||
/// With a conformed audio folder, reads come from the conformed file once it is written.
|
||||
pub fn audio_samples(&self, footage: &Footage, start: Tick, frames: usize, rate: u32) -> Vec<f32> {
|
||||
if !footage.has_audio || footage.missing || rate == 0 || frames == 0 {
|
||||
|
|
@ -828,7 +828,7 @@ impl FootageSource for MediaPool {
|
|||
|
||||
/// `frames` stereo sample frames (interleaved) of `src` from sample `s0` at `rate` Hz, which
|
||||
/// FilmCraft serves as decoded when it is the source's own rate. Mono is duplicated to both
|
||||
/// channels; silence before the start. An error where decoding fails.
|
||||
/// channels, multichannel sources are downmixed ([`crate::downmix`]); silence before the start. An error where decoding fails.
|
||||
fn read_stereo(src: &SharedSource, s0: i64, frames: usize, rate: u32) -> std::result::Result<Vec<f32>, String> {
|
||||
let mut out = vec![0.0; frames * 2];
|
||||
let skip = usize::try_from(s0.saturating_neg()).unwrap_or(0);
|
||||
|
|
@ -836,12 +836,9 @@ fn read_stereo(src: &SharedSource, s0: i64, frames: usize, rate: u32) -> std::re
|
|||
return Ok(out);
|
||||
}
|
||||
let buf = src.audio(s0.max(0), frames - skip, rate).map_err(|e| e.to_string())?;
|
||||
let (Some(l), Some(r)) = (buf.channels.first(), buf.channels.get(1).or(buf.channels.first())) else { return Ok(out) };
|
||||
let Some(dst) = skip.checked_mul(2).and_then(|i| out.get_mut(i..)) else { return Ok(out) };
|
||||
for ([o0, o1], (a, b)) in dst.as_chunks_mut::<2>().0.iter_mut().zip(l.iter().zip(r)) {
|
||||
*o0 = *a;
|
||||
*o1 = *b;
|
||||
}
|
||||
let aac = src.info().audio.as_ref().is_some_and(|a| a.codec.starts_with("AAC"));
|
||||
crate::downmix::downmix_into(&buf.channels, aac, dst);
|
||||
Ok(out)
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,226 @@
|
|||
//! Multichannel audio downmix (#903) against ffmpeg as an external oracle. Fixtures: ffmpeg with
|
||||
//! one sine per channel, written to `target/fixtures/media`; tests skip when ffmpeg is missing.
|
||||
|
||||
use std::f64::consts::PI;
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::process::Command;
|
||||
|
||||
use effectcraft_media::{MediaPool, probe};
|
||||
use effectcraft_time::Tick;
|
||||
|
||||
const RATE: u32 = 48_000;
|
||||
|
||||
fn fixtures() -> PathBuf {
|
||||
Path::new(env!("CARGO_MANIFEST_DIR")).join("../../target/fixtures/media")
|
||||
}
|
||||
|
||||
/// A fixture with `layout`'s channels `names` (join order) each carrying its own sine, encoded
|
||||
/// with `codec_args`. `None` when ffmpeg is missing or cannot make it.
|
||||
fn fixture(name: &str, layout: &str, names: &[&str], freqs: &[u32], codec_args: &[&str]) -> Option<PathBuf> {
|
||||
if !Command::new("ffmpeg").arg("-version").output().is_ok_and(|o| o.status.success()) {
|
||||
eprintln!("ffmpeg not found: skipping");
|
||||
return None;
|
||||
}
|
||||
let out = fixtures().join(name);
|
||||
if out.exists() {
|
||||
return Some(out);
|
||||
}
|
||||
std::fs::create_dir_all(fixtures()).ok()?;
|
||||
let ext = out.extension()?.to_string_lossy().to_string();
|
||||
let tmp = out.with_file_name(format!("tmp-{}-{name}", std::process::id())).with_extension(ext);
|
||||
let mut cmd = Command::new("ffmpeg");
|
||||
cmd.args(["-hide_banner", "-loglevel", "error", "-y"]);
|
||||
let mut map = String::new();
|
||||
for (i, (n, f)) in names.iter().zip(freqs).enumerate() {
|
||||
cmd.args(["-f", "lavfi", "-i", &format!("sine=frequency={f}:sample_rate={RATE}:duration=2")]);
|
||||
map.push_str(&format!("[{i}]"));
|
||||
let _ = n;
|
||||
}
|
||||
let post = if name.contains("loud") { ",volume=8:precision=float,alimiter=limit=1:level=disabled" } else { "" };
|
||||
let joins: Vec<String> = names.iter().enumerate().map(|(i, n)| format!("{i}.0-{n}")).collect();
|
||||
cmd.args(["-filter_complex", &format!("{map}join=inputs={}:channel_layout={layout}:map={}{}[a]", names.len(), joins.join("|"), post), "-map", "[a]"]);
|
||||
cmd.args(codec_args).arg(&tmp);
|
||||
let ok = cmd.status().ok()?.success();
|
||||
if !ok {
|
||||
eprintln!("ffmpeg could not make {name}");
|
||||
return None;
|
||||
}
|
||||
std::fs::rename(&tmp, &out).ok()?;
|
||||
Some(out)
|
||||
}
|
||||
|
||||
/// Amplitude (peak) of the sine at `freq` in one channel of interleaved stereo.
|
||||
fn amp(s: &[f32], ch: usize, freq: u32) -> f64 {
|
||||
let (mut re, mut im, mut n) = (0.0, 0.0, 0usize);
|
||||
for (i, f) in s.as_chunks::<2>().0.iter().enumerate() {
|
||||
let w = 2.0 * PI * f64::from(freq) * i as f64 / f64::from(RATE);
|
||||
re += f64::from(f[ch]) * w.cos();
|
||||
im += f64::from(f[ch]) * w.sin();
|
||||
n += 1;
|
||||
}
|
||||
2.0 * (re * re + im * im).sqrt() / n.max(1) as f64
|
||||
}
|
||||
|
||||
const K: f64 = std::f64::consts::FRAC_1_SQRT_2;
|
||||
|
||||
struct Case {
|
||||
file: &'static str,
|
||||
layout: &'static str,
|
||||
names: &'static [&'static str],
|
||||
codec: &'static [&'static str],
|
||||
/// Expected `[left, right]` gain of each channel (the same order as `names`).
|
||||
gains: Vec<[f64; 2]>,
|
||||
/// Allowed deviation of a channel's amplitude (lossy codecs).
|
||||
tol: f64,
|
||||
}
|
||||
|
||||
const FREQS: [u32; 8] = [300, 500, 700, 900, 1100, 1300, 1500, 1700];
|
||||
|
||||
fn five_one() -> (&'static [&'static str], Vec<[f64; 2]>) {
|
||||
let k = 1.0 / (1.0 + 2.0 * K);
|
||||
(&["FL", "FR", "FC", "LFE", "BL", "BR"], vec![[k, 0.0], [0.0, k], [k * K, k * K], [0.0, 0.0], [k * K, 0.0], [0.0, k * K]])
|
||||
}
|
||||
|
||||
fn seven_one() -> (&'static [&'static str], Vec<[f64; 2]>) {
|
||||
let k = 1.0 / (1.0 + 3.0 * K);
|
||||
(
|
||||
&["FL", "FR", "FC", "LFE", "BL", "BR", "SL", "SR"],
|
||||
vec![[k, 0.0], [0.0, k], [k * K, k * K], [0.0, 0.0], [k * K, 0.0], [0.0, k * K], [k * K, 0.0], [0.0, k * K]],
|
||||
)
|
||||
}
|
||||
|
||||
fn cases() -> Vec<Case> {
|
||||
let (n51, g51) = five_one();
|
||||
let (n71, g71) = seven_one();
|
||||
let k5 = 1.0 / (1.0 + 2.0 * K);
|
||||
let kq = 1.0 / (1.0 + K);
|
||||
vec![
|
||||
Case { file: "dm51.m4a", layout: "5.1", names: n51, codec: &["-c:a", "aac", "-b:a", "384k"], gains: g51.clone(), tol: 0.15 },
|
||||
Case { file: "dm51.ac3.mp4", layout: "5.1", names: n51, codec: &["-c:a", "ac3", "-b:a", "640k"], gains: g51.clone(), tol: 0.15 },
|
||||
Case { file: "dm51.wav", layout: "5.1", names: n51, codec: &["-c:a", "pcm_s16le"], gains: g51.clone(), tol: 0.02 },
|
||||
Case { file: "dm51.flac", layout: "5.1", names: n51, codec: &["-c:a", "flac"], gains: g51, tol: 0.02 },
|
||||
Case { file: "dm71.wav", layout: "7.1", names: n71, codec: &["-c:a", "pcm_s16le"], gains: g71.clone(), tol: 0.02 },
|
||||
Case { file: "dm71.flac", layout: "7.1", names: n71, codec: &["-c:a", "flac"], gains: g71, tol: 0.02 },
|
||||
Case {
|
||||
file: "dm50.wav",
|
||||
layout: "5.0",
|
||||
names: &["FL", "FR", "FC", "BL", "BR"],
|
||||
codec: &["-c:a", "pcm_s16le"],
|
||||
gains: vec![[k5, 0.0], [0.0, k5], [k5 * K, k5 * K], [k5 * K, 0.0], [0.0, k5 * K]],
|
||||
tol: 0.02,
|
||||
},
|
||||
Case {
|
||||
file: "dmquad.wav",
|
||||
layout: "quad",
|
||||
names: &["FL", "FR", "BL", "BR"],
|
||||
codec: &["-c:a", "pcm_s16le"],
|
||||
gains: vec![[kq, 0.0], [0.0, kq], [kq * K, 0.0], [0.0, kq * K]],
|
||||
tol: 0.02,
|
||||
},
|
||||
]
|
||||
}
|
||||
|
||||
/// Our stereo mix of `p` over one second, from 0.2 s in.
|
||||
fn ours(p: &Path) -> Vec<f32> {
|
||||
let f = probe(p).expect("probe");
|
||||
MediaPool::new().audio_samples(&f, Tick::from_seconds_f64(0.2), RATE as usize, RATE)
|
||||
}
|
||||
|
||||
/// ffmpeg's own `-ac 2` downmix of the same second, interleaved. Written as 16-bit integers:
|
||||
/// swresample only normalises the matrix (to keep full scale from clipping) for integer output.
|
||||
fn ffmpeg_ac2(p: &Path) -> Vec<f32> {
|
||||
let out = Command::new("ffmpeg")
|
||||
.args(["-hide_banner", "-loglevel", "error", "-ss", "0.2", "-t", "1", "-i"])
|
||||
.arg(p)
|
||||
.args(["-ac", "2", "-ar", "48000", "-f", "s16le", "-"])
|
||||
.output()
|
||||
.expect("run ffmpeg");
|
||||
out.stdout.as_chunks::<2>().0.iter().map(|b| f32::from(i16::from_le_bytes(*b)) / 32768.0).collect()
|
||||
}
|
||||
|
||||
fn rms(s: &[f32]) -> f64 {
|
||||
(s.iter().map(|v| f64::from(*v).powi(2)).sum::<f64>() / s.len().max(1) as f64).sqrt()
|
||||
}
|
||||
|
||||
fn peak(s: &[f32]) -> f64 {
|
||||
s.iter().fold(0.0f64, |m, v| m.max(f64::from(v.abs())))
|
||||
}
|
||||
|
||||
/// Each channel carries its own sine at amplitude 0.5·√... (ffmpeg's `sine` is −24 dBFS RMS,
|
||||
/// peak ≈ 0.125). Every channel must reach each side at the BS.775 gain, and the overall level
|
||||
/// must match ffmpeg's `-ac 2`.
|
||||
#[test]
|
||||
fn each_channel_reaches_stereo_at_its_bs775_gain() {
|
||||
for c in cases() {
|
||||
let freqs: Vec<u32> = FREQS.iter().copied().take(c.names.len()).collect();
|
||||
let Some(p) = fixture(c.file, c.layout, c.names, &freqs, c.codec) else { continue };
|
||||
let s = ours(&p);
|
||||
let sine = 0.125 * 1.0; // peak of ffmpeg's sine source (0.125 = 2^-3)
|
||||
for (i, (name, g)) in c.names.iter().zip(&c.gains).enumerate() {
|
||||
for (side, gain) in g.iter().enumerate() {
|
||||
let got = amp(&s, side, freqs[i]);
|
||||
let want = sine * gain;
|
||||
assert!((got - want).abs() <= sine * c.tol.max(0.02), "{}: {name} -> {} is {got:.4}, expected {want:.4}", c.file, ["L", "R"][side]);
|
||||
}
|
||||
}
|
||||
let theirs = ffmpeg_ac2(&p);
|
||||
let (a, b) = (rms(&s), rms(&theirs));
|
||||
let db = 20.0 * (a / b).log10();
|
||||
assert!(db.abs() < 1.0, "{}: level {db:.2} dB off ffmpeg -ac 2 (rms {a:.4} vs {b:.4})", c.file);
|
||||
let pdb = 20.0 * (peak(&s) / peak(&theirs)).log10();
|
||||
assert!(pdb.abs() < 1.5, "{}: peak {pdb:.2} dB off ffmpeg -ac 2", c.file);
|
||||
}
|
||||
}
|
||||
|
||||
/// Full-scale in every channel must not clip.
|
||||
#[test]
|
||||
fn loud_surround_does_not_clip() {
|
||||
let Some(p) = fixture("dm51_loud.wav", "5.1", &["FL", "FR", "FC", "LFE", "BL", "BR"], &[300; 6], &["-c:a", "pcm_s16le"]) else { return };
|
||||
let s = ours(&p);
|
||||
assert!(peak(&s) <= 1.001, "peak {}", peak(&s));
|
||||
assert!(peak(&s) > 0.5, "audible, peak {}", peak(&s));
|
||||
}
|
||||
|
||||
/// Mono and stereo footage keep their level (mono on both sides).
|
||||
#[test]
|
||||
fn mono_and_stereo_keep_their_level() {
|
||||
let Some(p) = fixture("dm10.wav", "mono", &["FC"], &[440], &["-c:a", "pcm_s16le"]) else { return };
|
||||
let s = ours(&p);
|
||||
assert!((amp(&s, 0, 440) - 0.125).abs() < 0.003 && (amp(&s, 1, 440) - 0.125).abs() < 0.003);
|
||||
let Some(p) = fixture("dm20.wav", "stereo", &["FL", "FR"], &[440, 660], &["-c:a", "pcm_s16le"]) else { return };
|
||||
let s = ours(&p);
|
||||
assert!((amp(&s, 0, 440) - 0.125).abs() < 0.003 && amp(&s, 1, 440) < 0.003);
|
||||
assert!((amp(&s, 1, 660) - 0.125).abs() < 0.003 && amp(&s, 0, 660) < 0.003);
|
||||
}
|
||||
|
||||
/// Odd channel counts (3, 4, 7, 9, 12, 16) decode to finite stereo without panicking.
|
||||
#[test]
|
||||
fn hostile_channel_counts_do_not_panic() {
|
||||
for (n, layout) in [(3, "3.0"), (4, "quad"), (7, "6.1"), (9, "9 channels"), (12, "12 channels"), (16, "16 channels")] {
|
||||
let names: Vec<String> = (0..n).map(|i| format!("{i}")).collect();
|
||||
// channel names for unnamed layouts are DEFAULT-ordered by position
|
||||
let _ = names;
|
||||
let file = format!("dmn{n}.wav");
|
||||
let args = ["-f", "lavfi", "-i", &format!("sine=frequency=440:sample_rate={RATE}:duration=1")];
|
||||
if !Command::new("ffmpeg").arg("-version").output().is_ok_and(|o| o.status.success()) {
|
||||
return;
|
||||
}
|
||||
let out = fixtures().join(&file);
|
||||
std::fs::create_dir_all(fixtures()).expect("mkdir");
|
||||
if !out.exists() {
|
||||
let tmp = fixtures().join(format!("tmp-{}-{file}", std::process::id()));
|
||||
let af = format!("pan={layout}|{}", (0..n).map(|i| format!("c{i}=c0")).collect::<Vec<_>>().join("|"));
|
||||
let ok =
|
||||
Command::new("ffmpeg").args(["-hide_banner", "-loglevel", "error", "-y"]).args(args).args(["-af", &af, "-c:a", "pcm_s16le"]).arg(&tmp).status();
|
||||
if !ok.is_ok_and(|s| s.success()) {
|
||||
eprintln!("ffmpeg cannot make {n} channels: skipping");
|
||||
continue;
|
||||
}
|
||||
std::fs::rename(&tmp, &out).expect("rename");
|
||||
}
|
||||
let Ok(f) = probe(&out) else { continue };
|
||||
let s = MediaPool::new().audio_samples(&f, Tick::ZERO, 4800, RATE);
|
||||
assert_eq!(s.len(), 9600, "{n} channels");
|
||||
assert!(s.iter().all(|v| v.is_finite() && v.abs() <= 1.001), "{n} channels");
|
||||
}
|
||||
}
|
||||
|
|
@ -615,9 +615,14 @@ fn flac_audio_of_any_shape_reads_in_device_blocks() {
|
|||
}
|
||||
at += 1024;
|
||||
}
|
||||
// ffmpeg's sine source has an amplitude of 1/8, on both channels (mono is duplicated).
|
||||
// ffmpeg's sine source has an amplitude of 1/8 (mono is duplicated). The fixture carries
|
||||
// channels 0 and 1 at full level and the others at half, so a multichannel file sounds as
|
||||
// its stereo downmix (tests/downmix.rs checks that matrix channel by channel).
|
||||
let gains = effectcraft_media::downmix::stereo_gains(channels as usize, false);
|
||||
let side = |s: usize| gains.iter().enumerate().map(|(c, g)| f64::from(g[s]) * if c < 2 { 1.0 } else { 0.5 }).sum::<f64>();
|
||||
let want = 0.125 * ((side(0).powi(2) + side(1).powi(2)) / 2.0).sqrt() / 2f64.sqrt();
|
||||
let rms = (sum / n.max(1) as f64).sqrt();
|
||||
assert!((rms - 0.125 / 2f64.sqrt()).abs() < 0.003, "{}: rms {rms}", p.display());
|
||||
assert!((rms - want).abs() < 0.003, "{}: rms {rms}, expected {want}", p.display());
|
||||
eprintln!("{}: read {secs} s in {:?}, slowest block after the first {worst:?}", p.display(), t0.elapsed());
|
||||
assert_eq!(pool.stats().opened, 1, "opened once, not per block");
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in New Issue