4 Commits

Author SHA1 Message Date
Bel LaPointe
fffea2ddf0 no render mac 2025-09-10 11:20:01 -06:00
Bel LaPointe
12dbf12299 k 2024-09-21 21:33:40 -04:00
Bel LaPointe
f04a55590f fixed 2024-09-21 21:33:40 -04:00
Bel LaPointe
2254afcbfb wav to mkv with subtitles scripting 2024-09-21 21:33:40 -04:00
6 changed files with 398 additions and 183 deletions

2
.gitmodules vendored
View File

@@ -1,3 +1,3 @@
[submodule "rust-whisper.d/gitea-whisper-rs"]
[submodule "gitea-whisper-rs"]
path = gitea-whisper-rs
url = https://gitea.inhome.blapointe.com/bel/whisper-rs.git

View File

@@ -1,236 +1,284 @@
use rust_whisper_lib;
use rust_whisper_baked_lib;
use clap::Parser;
use listen_lib;
use rust_whisper_baked_lib;
use rust_whisper_lib;
use std::thread;
fn main() {
let flags = rust_whisper_lib::Flags::parse();
match flags.wav.clone() {
Some(_) => wav_channel(flags),
None => channel(flags),
};
let flags = rust_whisper_lib::Flags::parse();
match flags.wav.clone() {
Some(_) => wav_channel(flags),
None => channel(flags),
};
}
fn wav_channel(flags: rust_whisper_lib::Flags) {
let mut w = new_destutterer();
rust_whisper_baked_lib::wav_channel(
flags.clone(),
move |result: Result<rust_whisper_lib::Transcribed, String>| {
match result {
Ok(transcribed) => {
let s = w.step(transcribed.to_string());
println!("{}", s);
},
Err(msg) => { eprintln!("error: {}", msg); },
};
},
);
let mut w = new_destutterer();
rust_whisper_baked_lib::wav_channel(
flags.clone(),
move |result: Result<rust_whisper_lib::Transcribed, String>| {
match result {
Ok(transcribed) => {
let s = w.step(transcribed.to_string());
println!("{}", s);
}
Err(msg) => {
eprintln!("error: {}", msg);
}
};
},
);
}
fn wav(flags: rust_whisper_lib::Flags, _path: String) {
let mut w = new_destutterer();
rust_whisper_baked_lib::wav(flags,
move |result: Result<rust_whisper_lib::Transcribed, String>| {
match result {
Ok(transcribed) => {
let s = w.step(transcribed.to_string());
println!("{}", s);
},
Err(msg) => { eprintln!("error: {}", msg); },
};
},
);
let mut w = new_destutterer();
rust_whisper_baked_lib::wav(
flags,
move |result: Result<rust_whisper_lib::Transcribed, String>| {
match result {
Ok(transcribed) => {
let s = w.step(transcribed.to_string());
println!("{}", s);
}
Err(msg) => {
eprintln!("error: {}", msg);
}
};
},
);
}
fn channel(flags: rust_whisper_lib::Flags) {
let (send, recv) = std::sync::mpsc::sync_channel(100);
let (send, recv) = std::sync::mpsc::sync_channel(100);
eprintln!("rust whisper baked lib channel...");
thread::spawn(move || {
let mut w = new_destutterer();
rust_whisper_baked_lib::channel(
flags.clone(),
move |result: Result<rust_whisper_lib::Transcribed, String>| {
match result {
Ok(transcribed) => {
let s = w.step(transcribed.to_string());
println!("{}", s);
},
Err(msg) => { eprintln!("error: {}", msg); },
};
},
recv,
);
});
eprintln!("rust whisper baked lib channel...");
thread::spawn(move || {
let mut w = new_destutterer();
rust_whisper_baked_lib::channel(
flags.clone(),
move |result: Result<rust_whisper_lib::Transcribed, String>| {
match result {
Ok(transcribed) => {
let s = w.step(transcribed.to_string());
println!("{}", s);
}
Err(msg) => {
eprintln!("error: {}", msg);
}
};
},
recv,
);
});
eprintln!("listen lib main...");
let flags = rust_whisper_lib::Flags::parse();
match flags.stream_device {
Some(device_name) => {
eprintln!("with device ({}) '{}'", device_name.len(), &device_name);
if device_name.len() == 0 {
let mut i = 0;
for device in listen_lib::devices() {
eprintln!("[{}] {}", i, device);
i += 1;
eprintln!("listen lib main...");
let flags = rust_whisper_lib::Flags::parse();
match flags.stream_device {
Some(device_name) => {
eprintln!("with device ({}) '{}'", device_name.len(), &device_name);
if device_name.len() == 0 {
let mut i = 0;
for device in listen_lib::devices() {
eprintln!("[{}] {}", i, device);
i += 1;
}
eprintln!("found {} devices", i);
} else {
listen_lib::main_with(
|data| {
send.send(data).unwrap();
},
device_name,
);
}
eprintln!("found {} devices", i);
} else {
listen_lib::main_with(|data| {
send.send(data).unwrap();
}, device_name);
}
},
None => {
eprintln!("without any device");
listen_lib::main(|data| {
send.send(data).unwrap();
});
}
}
eprintln!("/listen lib main...");
}
None => {
eprintln!("without any device");
listen_lib::main(|data| {
send.send(data).unwrap();
});
}
}
eprintln!("/listen lib main...");
}
struct Destutterer {
prev: Words,
prev: Words,
}
fn new_destutterer() -> Destutterer {
Destutterer{prev: new_words()}
Destutterer { prev: new_words() }
}
impl Destutterer {
fn step(&mut self, next: String) -> String {
if next.len() == 0 {
return next;
}
fn step(&mut self, next: String) -> String {
if next.len() == 0 {
return next;
}
let next_words = Words::from_string(next.clone());
let mut n = self.prev.comparable_len().clamp(0, next_words.comparable_len());
//println!("n={} prev='{:?}' next='{:?}'", n, self.prev.to_comparable_words(), next_words.to_comparable_words());
while n > 0 {
let (prev_s, _) = self.prev.last_n_comparable_to_string(n);
let (next_s, next_idx) = next_words.first_n_comparable_to_string(n);
if prev_s == next_s {
self.prev = next_words;
return self.prev.skip(next_idx+1).to_string();
}
n -= 1;
}
self.prev = next_words;
self.prev.to_string()
}
let next_words = Words::from_string(next.clone());
let mut n = self
.prev
.comparable_len()
.clamp(0, next_words.comparable_len());
//println!("n={} prev='{:?}' next='{:?}'", n, self.prev.to_comparable_words(), next_words.to_comparable_words());
while n > 0 {
let (prev_s, _) = self.prev.last_n_comparable_to_string(n);
let (next_s, next_idx) = next_words.first_n_comparable_to_string(n);
if prev_s == next_s {
self.prev = next_words;
return self.prev.skip(next_idx + 1).to_string();
}
n -= 1;
}
self.prev = next_words;
self.prev.to_string()
}
}
#[derive(Clone, Debug)]
struct Words {
raw: Vec<String>,
raw: Vec<String>,
}
fn new_words() -> Words {
Words{raw: vec![]}
Words { raw: vec![] }
}
impl Words {
fn from_string(s: String) -> Words {
let mut result = Words{raw: vec![]};
for word in s.split(" ") {
let word = word.trim();
if word.len() > 0 {
result.raw.push(word.to_string());
}
}
result
}
fn from_string(s: String) -> Words {
let mut result = Words { raw: vec![] };
for word in s.split(" ") {
let word = word.trim();
if word.len() > 0 {
result.raw.push(word.to_string());
}
}
result
}
fn skip(&self, n: usize) -> Words {
Words{
raw: self.raw.iter().skip(n).map(|x| x.clone()).collect(),
}
}
fn skip(&self, n: usize) -> Words {
Words {
raw: self.raw.iter().skip(n).map(|x| x.clone()).collect(),
}
}
fn last_n_comparable_to_string(&self, n: usize) -> (String, usize) {
let v = self.to_comparable_words();
let v = v[(v.len() - n).clamp(0, v.len())..].to_vec();
return (v.iter().map(|x| x.s.clone().unwrap()).collect::<Vec<String>>().join(" "), v[0].idx)
}
fn last_n_comparable_to_string(&self, n: usize) -> (String, usize) {
let v = self.to_comparable_words();
let v = v[(v.len() - n).clamp(0, v.len())..].to_vec();
return (
v.iter()
.map(|x| x.s.clone().unwrap())
.collect::<Vec<String>>()
.join(" "),
v[0].idx,
);
}
fn first_n_comparable_to_string(&self, n: usize) -> (String, usize){
let v = self.to_comparable_words();
let v = v[0..n.clamp(0, v.len())].to_vec();
return (v.iter().map(|x| x.s.clone().unwrap()).collect::<Vec<String>>().join(" "), v[v.len()-1].idx)
}
fn first_n_comparable_to_string(&self, n: usize) -> (String, usize) {
let v = self.to_comparable_words();
let v = v[0..n.clamp(0, v.len())].to_vec();
return (
v.iter()
.map(|x| x.s.clone().unwrap())
.collect::<Vec<String>>()
.join(" "),
v[v.len() - 1].idx,
);
}
fn comparable_len(&self) -> usize {
self.to_comparable_words().len()
}
fn comparable_len(&self) -> usize {
self.to_comparable_words().len()
}
fn to_comparable_words(&self) -> Vec<Word> {
self.to_words().iter()
.filter(|x| x.s.is_some())
.map(|x| x.clone())
.collect()
}
fn to_comparable_words(&self) -> Vec<Word> {
self.to_words()
.iter()
.filter(|x| x.s.is_some())
.map(|x| x.clone())
.collect()
}
fn to_words(&self) -> Vec<Word> {
let skips = stop_words::get("en");
let stemmer = rust_stemmers::Stemmer::create(rust_stemmers::Algorithm::English);
let strs = self.raw.iter()
.map(|w| w.to_lowercase())
.map(|w| w.chars().filter(|c| c.is_ascii_alphanumeric()).collect::<String>())
.map(|w| stemmer.stem(&w).into_owned())
.collect::<Vec<String>>();
let mut result = vec![];
for i in 0..strs.len() {
result.push(Word{
s: match skips.contains(&strs[i]) {
true => None,
false => Some(strs[i].clone()),
},
idx: i as usize,
});
}
result
}
fn to_words(&self) -> Vec<Word> {
let skips = stop_words::get("en");
let stemmer = rust_stemmers::Stemmer::create(rust_stemmers::Algorithm::English);
let strs = self
.raw
.iter()
.map(|w| w.to_lowercase())
.map(|w| {
w.chars()
.filter(|c| c.is_ascii_alphanumeric())
.collect::<String>()
})
.map(|w| stemmer.stem(&w).into_owned())
.collect::<Vec<String>>();
let mut result = vec![];
for i in 0..strs.len() {
result.push(Word {
s: match skips.contains(&strs[i]) {
true => None,
false => Some(strs[i].clone()),
},
idx: i as usize,
});
}
result
}
fn to_string(&self) -> String {
self.raw.iter()
.map(|x| x.clone())
.collect::<Vec<String>>()
.join(" ")
}
fn to_string(&self) -> String {
self.raw
.iter()
.map(|x| x.clone())
.collect::<Vec<String>>()
.join(" ")
}
}
#[derive(Debug, Clone)]
struct Word {
s: Option<String>,
idx: usize,
s: Option<String>,
idx: usize,
}
#[cfg(test)]
mod tests {
use super::*;
use super::*;
#[test]
fn test_destutterer_stop_words() {
let mut w = new_destutterer();
assert_eq!("welcome to the internet".to_string(), w.step("welcome to the internet".to_string()));
assert_eq!("have a look around".to_string(), w.step("welcome to the a internet; have a look around".to_string()));
}
#[test]
fn test_destutterer_stop_words() {
let mut w = new_destutterer();
assert_eq!(
"welcome to the internet".to_string(),
w.step("welcome to the internet".to_string())
);
assert_eq!(
"have a look around".to_string(),
w.step("welcome to the a internet; have a look around".to_string())
);
}
#[test]
fn test_destutterer_punctuation() {
let mut w = new_destutterer();
assert_eq!("cat, dog. cow? moose!".to_string(), w.step("cat, dog. cow? moose!".to_string()));
assert_eq!("elephant! fez gator".to_string(), w.step("moose, elephant! fez gator".to_string()));
assert_eq!("hij".to_string(), w.step("fez gator hij".to_string()));
}
#[test]
fn test_destutterer_punctuation() {
let mut w = new_destutterer();
assert_eq!(
"cat, dog. cow? moose!".to_string(),
w.step("cat, dog. cow? moose!".to_string())
);
assert_eq!(
"elephant! fez gator".to_string(),
w.step("moose, elephant! fez gator".to_string())
);
assert_eq!("hij".to_string(), w.step("fez gator hij".to_string()));
}
#[test]
fn test_destutterer_basic() {
let mut w = new_destutterer();
assert_eq!("cat dog cow".to_string(), w.step(" cat dog cow ".to_string()));
assert_eq!("moose".to_string(), w.step(" dog cow moose ".to_string()));
}
#[test]
fn test_destutterer_basic() {
let mut w = new_destutterer();
assert_eq!(
"cat dog cow".to_string(),
w.step(" cat dog cow ".to_string())
);
assert_eq!("moose".to_string(), w.step(" dog cow moose ".to_string()));
}
}

View File

@@ -1,4 +1,5 @@
todo:
- wav to subtitles
- compound words like checkmark vs check mark should destutter
- whisper trims outside silence so head and tail never get hit
- split on silence-ish instead of duration
@@ -6,6 +7,6 @@ todo:
scheduled: []
done:
- todo: need to overlap without ANY puctuation, which i can do by breaking into words
ts: Tue Jan 2 18:23:00 MST 2024
ts: Tue Jan 2 13:23:00 EST 2024
- todo: overlap without stop words
ts: Wed Jan 3 08:22:14 MST 2024
ts: Wed Jan 3 03:22:14 EST 2024

BIN
wav_to_mkv.d/sc.jpg Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 11 KiB

View File

@@ -0,0 +1,100 @@
// This example is not going to build in this folder.
// You need to copy this code into your project and add the dependencies whisper_rs and hound in your cargo.toml
use hound;
use std::fs::File;
use std::io::Write;
use whisper_rs::{FullParams, SamplingStrategy, WhisperContext};
/// Loads a context and model, processes an audio file, and prints the resulting transcript to stdout.
fn main() -> Result<(), &'static str> {
let args: Vec<String> = std::env::args().collect();
// Load a context and model.
let ctx = WhisperContext::new(&args[1])
.expect("failed to load model");
// Create a state
let mut state = ctx.create_state().expect("failed to create key");
// Create a params object for running the model.
// The number of past samples to consider defaults to 0.
let mut params = FullParams::new(SamplingStrategy::Greedy { best_of: 0 });
// Edit params as needed.
// Set the number of threads to use to 1.
//params.set_n_threads(1);
// Enable translation.
params.set_translate(true);
// Set the language to translate to to English.
params.set_language(Some("en"));
// Disable anything that prints to stdout.
params.set_print_special(false);
params.set_print_progress(false);
params.set_print_realtime(false);
params.set_print_timestamps(false);
// Open the audio file.
let mut reader = hound::WavReader::open(&args[2]).expect("failed to open file");
#[allow(unused_variables)]
let hound::WavSpec {
channels,
sample_rate,
bits_per_sample,
..
} = reader.spec();
// Convert the audio to floating point samples.
let mut audio = whisper_rs::convert_integer_to_float_audio(
&reader
.samples::<i16>()
.map(|s| s.expect("invalid sample"))
.collect::<Vec<_>>(),
);
// Convert audio to 16KHz mono f32 samples, as required by the model.
// These utilities are provided for convenience, but can be replaced with custom conversion logic.
// SIMD variants of these functions are also available on nightly Rust (see the docs).
if channels == 2 {
audio = whisper_rs::convert_stereo_to_mono_audio(&audio)?;
} else if channels != 1 {
panic!(">2 channels unsupported");
}
if sample_rate != 16000 {
panic!("sample rate must be 16KHz");
}
// Run the model.
state.full(params, &audio[..]).expect("failed to run model");
// Create a file to write the transcript to.
let mut file = File::create("transcript.txt").expect("failed to create file");
// Iterate through the segments of the transcript.
let num_segments = state
.full_n_segments()
.expect("failed to get number of segments");
for i in 0..num_segments {
// Get the transcribed text and timestamps for the current segment.
let segment = state
.full_get_segment_text(i)
.expect("failed to get segment");
let start_timestamp = state
.full_get_segment_t0(i)
.expect("failed to get start timestamp");
let end_timestamp = state
.full_get_segment_t1(i)
.expect("failed to get end timestamp");
// Print the segment to stdout.
println!("[{} - {}]: {}", start_timestamp, end_timestamp, segment);
// Format the segment information as a string.
let line = format!("[{} - {}]: {}\n", start_timestamp, end_timestamp, segment);
// Write the segment information to the file.
file.write_all(line.as_bytes())
.expect("failed to write to file");
}
Ok(())
}

View File

@@ -0,0 +1,66 @@
#! /bin/bash
main() {
set -euo pipefail
input_wav="$(realpath "$1")"
model="$(realpath "${2:-../models/ggml-small.en.bin}")"
already_transcribed="${3:-false}"
sanitized_wav="${input_wav%.*}.mono-16khz.wav"
ffmpeg -y -i "$input_wav" -ac 1 -ar 16k "$sanitized_wav"
if ! $already_transcribed; then
pushd "$(dirname "$(realpath "$BASH_SOURCE")")"
cd ../gitea-whisper-rs/
cargo run --example wav_subtitles -- "$model" "$sanitized_wav"
popd
fi
out_to_srt ../gitea-whisper-rs/transcript.txt > "${input_wav%.*}.srt"
ffmpeg -y \
-loop 1 -i sc.jpg \
-i "$input_wav" \
-i "${input_wav%.*}.srt" \
-c:v libx264 \
-tune stillimage \
-pix_fmt yuv420p -shortest \
"${input_wav%.*}.mkv"
ls "${input_wav%.*}.mkv"
}
out_to_srt() {
cs_to_ts() {
echo "$1" | awk '{
printf "%02d:%02d:%02d,000",
int(($1/100.0)/60/60),
int(($1/100.0)/60%60),
int(($1/100.0)%60)
}'
}
cat "$1" \
| (
i=0
while read -r line; do
((i+=1))
echo "$i"
echo "$(cs_to_ts "$(
echo "${line%%:] *}" \
| tr -d '[' \
| awk '{print $1}'
)") --> $(cs_to_ts "$(
echo "${line%%:] *}" \
| tr -d '[' \
| awk '{print $3}'
)")"
echo "${line#*: }"
echo
done
)
}
if [ "$0" == "$BASH_SOURCE" ]; then
main "$@"
fi