Re-encode plain WAV sources with an unspecified channel layout

CHANGELOG.md

@@ -1,5 +1,17 @@
 # Changelog
 
+## Unreleased
+
+### Fixed
+
+- `extract_audio/3` and `transcode/3` re-encode a plain mono or stereo
+  WAV again. Before, every call that resampled or re-encoded such a file,
+  for example to `.mp3`, `.m4a`, or `.opus`, returned
+  `{:error, %Exmpeg.Error{reason: :io_error}}` with the message
+  ``AVERROR(-1668179713): `Input changed` ``. The WAV leaves its channel
+  layout unspecified, and the decoder now gets the default layout for its
+  channel count, as in the `ffmpeg` CLI.
+
 ## 0.6.0 - 2026-09-23
 
 ### Changed

native/exmpeg_native/src/audio.rs

@@ -1,8 +1,9 @@
 //! Shared helpers for the audio re-encode paths (`extract_audio`,
 //! `transcode`).
 
-use rsmpeg::avcodec::AVCodecRef;
+use rsmpeg::avcodec::{AVCodecContext, AVCodecRef};
 use rsmpeg::avutil::{AVChannelLayout, AVFrame};
+use rsmpeg::ffi;
 
 use crate::errors::NativeError;
 
@@ -35,6 +36,18 @@ pub(crate) fn resolve_channels(requested: Option<i32>, src: i32) -> Result<i32,
     Ok(target)
 }
 
+/// Give a decoder whose channel order is unspecified (a plain WAV) the
+/// default layout for its channel count, as the ffmpeg CLI does. Call it
+/// between `apply_codecpar` and `open`. Otherwise `swr_init` normalises
+/// its input layout to that default, and `swr_convert_frame` rejects
+/// every decoded frame with `AVERROR_INPUT_CHANGED`.
+pub(crate) fn default_unspecified_layout(decoder: &mut AVCodecContext) {
+    if decoder.ch_layout.order == ffi::AV_CHANNEL_ORDER_UNSPEC {
+        let layout = AVChannelLayout::from_nb_channels(decoder.ch_layout.nb_channels);
+        decoder.set_ch_layout(layout.into_inner());
+    }
+}
+
 pub(crate) fn pick_sample_fmt(codec: &AVCodecRef<'static>, src: i32) -> i32 {
     if let Some(fmts) = codec.sample_fmts() {
         if fmts.contains(&src) {

native/exmpeg_native/src/extract_audio.rs

@@ -78,6 +78,7 @@ pub(crate) fn extract_audio<Q: AsRef<Path>>(
         let codecpar = input.streams()[audio_index].codecpar();
         decoder.apply_codecpar(&codecpar)?;
     }
+    crate::audio::default_unspecified_layout(&mut decoder);
     decoder.set_time_base(stream_time_base);
     decoder.open(None)?;
 
@@ -109,18 +110,13 @@ pub(crate) fn extract_audio<Q: AsRef<Path>>(
     // The layout must match for this to be safe: a 2-channel source with
     // a *specific* non-stereo mask (e.g. WAVEFORMATEXTENSIBLE FL+LFE)
     // sent unremapped under the canonical target layout the output
-    // stream advertises would mislabel/misorder its channels. An
-    // *unspecified* source order with the matching channel count has no
-    // order to preserve, so passthrough under the canonical target is
-    // fine (and avoids the resampler for the common WAV case, which
-    // reports its layout as unspecified).
-    let layout_ok = ffi_helpers::channel_layouts_equal(&decoder.ch_layout, &target_layout)
-        || (decoder.ch_layout.order == ffi::AV_CHANNEL_ORDER_UNSPEC
-            && decoder.ch_layout.nb_channels == target_channels);
+    // stream advertises would mislabel/misorder its channels. A plain
+    // WAV with an unspecified order already has the default layout from
+    // `default_unspecified_layout`, so it matches here.
     let passthrough = encoder.frame_size == 0
         && decoder.sample_fmt == target_fmt
         && decoder.sample_rate == target_rate
-        && layout_ok;
+        && ffi_helpers::channel_layouts_equal(&decoder.ch_layout, &target_layout);
 
     let mut swr = if passthrough {
         None

native/exmpeg_native/src/transcode.rs

@@ -645,6 +645,7 @@ fn build_audio_pipeline(
     })?;
     let mut decoder = AVCodecContext::new(&decoder_codec);
     decoder.apply_codecpar(codecpar)?;
+    crate::audio::default_unspecified_layout(&mut decoder);
     // The decoder moves the timestamps of a frame it trims (Opus and AAC
     // priming) only when it knows the packet time_base.
     decoder.set_pkt_timebase(in_tb);

test/integration_test.exs

@@ -971,6 +971,35 @@ defmodule Exmpeg.IntegrationTest do
     end
   end
 
+  # The audio-only fixture is a plain mono WAV, so its channel layout is
+  # unspecified. The next three tests re-encode it.
+  test "extract_audio re-encodes a mono 44.1 kHz WAV to mp3" do
+    wav = TestFixtures.ensure_audio_only_clip!()
+    out = Path.join(System.tmp_dir!(), "exmpeg_wav_mp3_#{System.unique_integer([:positive])}.mp3")
+    on_exit(fn -> File.rm(out) end)
+
+    assert {:ok, _stats} = Exmpeg.extract_audio(wav, out)
+    assert {:ok, %MediaInfo{streams: [audio]}} = Exmpeg.probe(out)
+    assert audio.codec in ["mp3", "mp3float"]
+    assert %{sample_rate: 44_100, channels: 1} = audio.audio
+  end
+
+  test "extract_audio resamples a mono 44.1 kHz WAV to 48 kHz opus" do
+    wav = TestFixtures.ensure_audio_only_clip!()
+    out = Path.join(System.tmp_dir!(), "exmpeg_wav_opus_#{System.unique_integer([:positive])}.opus")
+    on_exit(fn -> File.rm(out) end)
+
+    assert {:ok, %{sample_rate: 48_000}} = Exmpeg.extract_audio(wav, out)
+  end
+
+  test "transcode re-encodes a mono 44.1 kHz WAV to aac in m4a" do
+    wav = TestFixtures.ensure_audio_only_clip!()
+    out = Path.join(System.tmp_dir!(), "exmpeg_wav_aac_#{System.unique_integer([:positive])}.m4a")
+    on_exit(fn -> File.rm(out) end)
+
+    assert {:ok, %{streams_reencoded: 1}} = Exmpeg.transcode(wav, out, audio_codec: "aac")
+  end
+
   test "concat rejects inputs with mismatched stream layouts", %{clip: clip} do
     video_only = TestFixtures.ensure_video_only_clip!()
     out = Path.join(System.tmp_dir!(), "exmpeg_concat_bad_#{System.unique_integer([:positive])}.mp4")