Add local spoken passages with Google retrieval and isolated runtime setup
This commit is contained in:
karl committed
1 parent
bc9459785d
commit
dab703413e
55 files changed
+3633
-70
No files matched your search
@@ -0,0 +1,225 @@
|
||||
//! Timeline contract uses actual CPU FFmpeg, without speech/model inference.
|
||||
use anyhow::{Context, Result};
|
||||
use searchi_media::{decode_audio, decode_audio_aligned, probe, AudioWindow};
|
||||
use searchi_service::{FfmpegTools, MediaLimits};
|
||||
use std::{path::Path, process::Command};
|
||||
|
||||
fn fixture(ffmpeg: &Path, args: &[&str], output: &Path) -> Result<()> {
|
||||
let result = Command::new(ffmpeg)
|
||||
.args([
|
||||
"-nostdin",
|
||||
"-hide_banner",
|
||||
"-loglevel",
|
||||
"error",
|
||||
"-threads",
|
||||
"1",
|
||||
])
|
||||
.args(args)
|
||||
.arg(output)
|
||||
.output()?;
|
||||
anyhow::ensure!(
|
||||
result.status.success(),
|
||||
"Owned CPU media fixture failed: {}",
|
||||
String::from_utf8_lossy(&result.stderr)
|
||||
);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[ignore = "Requires verified CPU FFmpeg/FFprobe helpers; no model inference"]
|
||||
fn positioned_audio_decodes_at_its_actual_file_timeline_origin() -> Result<()> {
|
||||
assert_eq!(std::env::var("SEARCHI_FORCE_CPU")?, "1");
|
||||
let tools = FfmpegTools {
|
||||
ffmpeg: std::env::var_os("SEARCHI_TEST_FFMPEG")
|
||||
.context("Set SEARCHI_TEST_FFMPEG")?
|
||||
.into(),
|
||||
ffprobe: std::env::var_os("SEARCHI_TEST_FFPROBE")
|
||||
.context("Set SEARCHI_TEST_FFPROBE")?
|
||||
.into(),
|
||||
};
|
||||
let directory = tempfile::tempdir()?;
|
||||
let ordinary = directory.path().join("ordinary.wav");
|
||||
let positioned = directory.path().join("positioned.mkv");
|
||||
fixture(
|
||||
&tools.ffmpeg,
|
||||
&[
|
||||
"-f",
|
||||
"lavfi",
|
||||
"-i",
|
||||
"sine=frequency=440:sample_rate=16000:duration=2",
|
||||
"-c:a",
|
||||
"pcm_s16le",
|
||||
],
|
||||
&ordinary,
|
||||
)?;
|
||||
fixture(
|
||||
&tools.ffmpeg,
|
||||
&[
|
||||
"-f",
|
||||
"lavfi",
|
||||
"-i",
|
||||
"color=size=16x16:rate=1:duration=4",
|
||||
"-itsoffset",
|
||||
"2",
|
||||
"-f",
|
||||
"lavfi",
|
||||
"-i",
|
||||
"sine=frequency=440:sample_rate=16000:duration=2",
|
||||
"-map",
|
||||
"0:v:0",
|
||||
"-map",
|
||||
"1:a:0",
|
||||
"-c:v",
|
||||
"ffv1",
|
||||
"-c:a",
|
||||
"pcm_s16le",
|
||||
"-threads",
|
||||
"1",
|
||||
],
|
||||
&positioned,
|
||||
)?;
|
||||
let limits = MediaLimits::default();
|
||||
let metadata = probe(&tools, &positioned, &limits)?;
|
||||
assert_eq!(metadata.start_ms, 0);
|
||||
assert_eq!(metadata.audio_start_ms, 2000);
|
||||
assert_eq!(metadata.audio_end_ms, 4000);
|
||||
let baseline = decode_audio(
|
||||
&tools,
|
||||
&ordinary,
|
||||
AudioWindow {
|
||||
start_ms: 0,
|
||||
duration_ms: 2000,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
let actual = decode_audio(
|
||||
&tools,
|
||||
&positioned,
|
||||
AudioWindow {
|
||||
start_ms: metadata.audio_start_ms,
|
||||
duration_ms: metadata.audio_end_ms - metadata.audio_start_ms,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(baseline.len(), 32000);
|
||||
let aligned = decode_audio_aligned(
|
||||
&tools,
|
||||
&positioned,
|
||||
AudioWindow {
|
||||
start_ms: 2000,
|
||||
duration_ms: 2000,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(aligned, baseline);
|
||||
let whole = decode_audio_aligned(
|
||||
&tools,
|
||||
&positioned,
|
||||
AudioWindow {
|
||||
start_ms: 0,
|
||||
duration_ms: 4000,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(whole.len(), 64000);
|
||||
assert!(whole[..32000].iter().all(|sample| *sample == 0.));
|
||||
assert_eq!(&whole[32000..], baseline.as_slice());
|
||||
assert_eq!(
|
||||
actual, baseline,
|
||||
"First decoded sample must represent the requested actual audio origin"
|
||||
);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[ignore = "Requires verified CPU FFmpeg/FFprobe helpers; no model inference"]
|
||||
fn aligned_decoder_preserves_internal_gap_seek_and_tail_source_clock() -> Result<()> {
|
||||
assert_eq!(std::env::var("SEARCHI_FORCE_CPU")?, "1");
|
||||
let tools = FfmpegTools {
|
||||
ffmpeg: std::env::var_os("SEARCHI_TEST_FFMPEG")
|
||||
.context("Set SEARCHI_TEST_FFMPEG")?
|
||||
.into(),
|
||||
ffprobe: std::env::var_os("SEARCHI_TEST_FFPROBE")
|
||||
.context("Set SEARCHI_TEST_FFPROBE")?
|
||||
.into(),
|
||||
};
|
||||
let directory = tempfile::tempdir()?;
|
||||
let ordinary = directory.path().join("ramp.wav");
|
||||
let gapped = directory.path().join("gapped.mka");
|
||||
fixture(
|
||||
&tools.ffmpeg,
|
||||
&[
|
||||
"-f",
|
||||
"lavfi",
|
||||
"-i",
|
||||
"aevalsrc=0.1+0.1*t:s=16000:d=3",
|
||||
"-c:a",
|
||||
"pcm_s16le",
|
||||
],
|
||||
&ordinary,
|
||||
)?;
|
||||
// Three exact one-second frames carry PTS 0,2,3; the source clock has a one-second gap.
|
||||
fixture(
|
||||
&tools.ffmpeg,
|
||||
&[
|
||||
"-f",
|
||||
"lavfi",
|
||||
"-i",
|
||||
"aevalsrc=0.1+0.1*t:s=16000:d=3",
|
||||
"-af",
|
||||
"asetnsamples=n=16000:p=0,asetpts=PTS+gte(T\\,1)/TB",
|
||||
"-c:a",
|
||||
"pcm_s16le",
|
||||
],
|
||||
&gapped,
|
||||
)?;
|
||||
let limits = MediaLimits::default();
|
||||
let metadata = probe(&tools, &gapped, &limits)?;
|
||||
assert_eq!(metadata.start_ms, 0);
|
||||
assert_eq!(metadata.audio_start_ms, 0);
|
||||
assert_eq!(metadata.audio_end_ms, 4000);
|
||||
let source = decode_audio(
|
||||
&tools,
|
||||
&ordinary,
|
||||
AudioWindow {
|
||||
start_ms: 0,
|
||||
duration_ms: 3000,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(source.len(), 48000);
|
||||
let mut expected = source[..16000].to_vec();
|
||||
expected.extend(vec![0.; 16000]);
|
||||
expected.extend_from_slice(&source[16000..]);
|
||||
let whole = decode_audio_aligned(
|
||||
&tools,
|
||||
&gapped,
|
||||
AudioWindow {
|
||||
start_ms: 0,
|
||||
duration_ms: 4000,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(whole, expected);
|
||||
let seek = decode_audio_aligned(
|
||||
&tools,
|
||||
&gapped,
|
||||
AudioWindow {
|
||||
start_ms: 1500,
|
||||
duration_ms: 1500,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(seek, &expected[24000..48000]);
|
||||
let tail = decode_audio_aligned(
|
||||
&tools,
|
||||
&gapped,
|
||||
AudioWindow {
|
||||
start_ms: 3500,
|
||||
duration_ms: 500,
|
||||
},
|
||||
&limits,
|
||||
)?;
|
||||
assert_eq!(tail, &expected[56000..64000]);
|
||||
Ok(())
|
||||
}
|
||||
Reference in new issue
Block a user