So I'm trying to write a simple program that will detect voice activity with a .wav file using the CMU Sphinx library.
So far, I have the following
SpeechClassifier s = new SpeechClassifier();
s.setPredecessor(dataSource);
Data d = s.getData();
while(d != null) {
if(s.isSpeech()) {
System.out.println("Speech is detected");
}
else {
System.out.println("Speech has not been detected");
}
System.out.println();
d = s.getData();
}
I get the output "Speech is not detected" but there is Speech in the audio file. It seems as if the getData function is not working the way I want it to. I want it to get the frames and then determine whether the frames (s.isSpeech()) contain speech or not.
I'm trying to have multiple outputs ("Speech is detected" vs "Speech is not detected") for each frame. How can I make my code better? Thanks!
You need to insert DataBlocker before SpeechClassifier:
DataBlocker b = new DataBlocker(10); // means 10ms
SpeechClassifier s = new SpeechClassifier(10, 0.003, 10, 0);
b.setPredecessor(dataSource);
s.setPredecessor(b);
Then it will process 10 millisecond frames.
Related
I have this for statement triggered by a button that iterates through a MutableList of strings.
For each string it completes a file path and checks if that file path is valid. If it is, it's attempted to be sent to the mediaPlayer via a function to be played as a sound file. It should play the sound for all files it can find with a pause at certain stated points (2,5,7).
Unfortunately, when I test it out, the button animation comes in delayed, followed by a Logcat info of 363 skipped frames, doing too much work on the application's main thread.
I tried to consecutively commenting out certain lines of the function but was not able to identify the computationally intensive part of it. Could anybody tell me where the issue lies or how I could improve the function?
Here's the function itself:
btnStartReadingAloud.setOnClickListener {binding.root.context
Toast.makeText(binding.root.context, "Reading the exercise out loud", Toast.LENGTH_SHORT).show()
println("Reading the exercise out loud")
for (i in 0 until newExercise.lastIndex){
val currentElement : Pair<String,Array<Any>> = newExercise[i]
val currentDesignatedSoundFile : String = "R.id.trampolin_ansage_malte_"+replaceSpecialChars(currentElement.first)
val path : Uri = Uri.parse(currentDesignatedSoundFile)
val file = File(currentDesignatedSoundFile)
if (doesFileExist(file)){
System.out.println("Playing file named$file")
playSound(path)
Toast.makeText(binding.root.context, "Playing sound for: %path", Toast.LENGTH_SHORT).show()
}
val pauseTimes = listOf<Int>(2,5,7)
if (i in pauseTimes){
Thread.sleep(2000)
}
}
Here is the dedicated function to play the sound
fun playSound(soundFile : Uri) {
if (mMediaPlayer == null) {
mMediaPlayer = MediaPlayer.create(requireContext(), soundFile)
mMediaPlayer!!.isLooping = false
mMediaPlayer!!.start()
} else mMediaPlayer!!.start()
}
Any hint is appreciated, thanks already for reading & brainstorming :)
Lets say that I am reading from data stream, and that stream is sending the content of an h264 video feed. Given I read from that stream and I have some amount of data consisting of an indeterminate number of frames (NAL?). Given that i know the framerate, and size of the originating video, how would I go about converting this snippet into a mp4 that i could view? The video does not contain audio.
I want to do this using nodejs? My attempts to do so have produced nothing resembling a valid h264 file to convert into mp4. My thoughts so far were to strip any data preceding the first found start code in the data and feed that into a file and use ffmpeg (currently just testing in the command line) to convert the file to mp4.
What's the correct way to go about doing this?
ie. something like this (it's in Typescript but same thing)
//We assume here that when this while loop exist at least one full frame of data will have been read and written to disk
let stream: WriteStream = fs.createWriteStream("./test.h264")
while(someDataStream.available()) { //just an example not real code
let data: Buffer = someDataStream.readSomeData() //just an example not a real method call
let file = null;
try {
file = fs.statSync("./test.h264");
} catch (error) {
console.error(error)
}
if(!stream.writable) {
console.error("stream not writable")
} else if(file == null || file.size <= 0) {
let index = data.indexOf(0x7C)
console.log("index: " + index)
if(index > 0) {
console.log("index2: " + data.slice(index).indexOf(0x7c))
stream.write(data.slice(index))
}
} else {
stream.write(data)
}
}
To handle a data stream, you'll need to emit fragmented MP4. Like all MP4, fMP4 streams begin with a preamble containing ftyp, moov, and styp boxes. Then each frame is encoded with a moof / mdat box pair.
In order to generate a useful preamble from your H.264 bitstream, you need to locate a SPS / PPS pair of NALUs in the H264 data, to set up the avc1 box within the moov box. Those two NALUs are often immediately followed by an I-frame (a key frame). The first frame in a stream must be an I-frame, and subsequent ones can be P- or B- frames. E
It's a fairly complex task involving lots of bit-banging and buffer-shuffling (those are technical terms ;-).
I've been working on a piece of js code to extract H.264 from webm and put it into fmp4. It's not yet complete. It's backed up by another piece of code to decode the parts of the H264 stream that are needed to pack it properly into fMP4.
I wish I could write, "here are the ten lines of code you need" but those formats (fMP4 and H264) aren't simple enough to make that possible.
Idk why none of those questions doesn't actually have an easy answer. Here you go, Node.js solution, i argument just in case you need to offset the search
const soi = Buffer.from([0x00, 0x00, 0x00, 0x01]);
function findStartFrame(buffer, i = -1) {
while ((i = buffer.indexOf(soi, i + 1)) !== -1) {
if ((buffer[i + 4] & 0x1F) === 7) return i
}
return -1
}
I need to concatenate two wav audio files with 30 seconds of whute sound between them.
I want to use the NAudio library - or with any other way that work.
How to do it ?
( the different from any other question is that i need not only to make one audio file from two different audio files .. i also need to add silent between them )
Assuming your WAV files have the same sample rate and channel count, you can concatenate using FollowedBy and use SignalGenerator combined with Take to get the white noise.
var f1 = new AudioFileReader("ex1.wav");
var f2 = new SignalGenerator(f1.WaveFormat.SampleRate, f1.WaveFormat.Channels) { Type = SignalGeneratorType.White, Gain = 0.2f }.Take(TimeSpan.FromSeconds(5));
var f3 = new AudioFileReader("ex3.wav");
using (var wo = new WaveOutEvent())
{
wo.Init(f1.FollowedBy(f2).FollowedBy(f3));
wo.Play();
while (wo.PlaybackState == PlaybackState.Playing) Thread.Sleep(500);
}
I have run into a little problem. Basicly i want to exstract from String-data off a pdf file.
More specifik this pdf file
http://www.midttrafik.dk/koereplaner/bybusser/aarhus/bybusser-aarhus/18-mejlbyelev-park-all%C3%A9-skaade-moesgaard/koereplan
So, my problem lays in not knowing, how to get the names, and the times(the pdf is times and locations of bus-stops, street names on the left kolon, and bus ariving times is the rest). the info i want to save is the number befor the street name (1-4), the street name, and all of the times.
translate of some of the stuff on the pdf.
Faste minuttal - just means that bus times is the same for the intival under 'Faste
6.56 - 8.11 - this means that, in this intival followes the under.
so
the buss will stop at 'Elev Skole, Høvej' 56, 11, 26, 41 meaning 6.56, 7.11, 7.26, 7.41, 7.56, 8.11.
I dont think i can desribe my problem any better, so i hope one of you guys will be able to help. i dont need a ready code, just send me in the rigth direaction - tell me what i can do, that migth help, or good patterns to use.
Thanks
You can use the nice PDFBox Library from here to extract the text you want from this pdf file. It works really nice, i used it in one of my last projects to index pfd files for a full text search.
Here is the URL to the project:
http://pdfbox.apache.org/index.html
There you'll find also the documentation and some examples how to extract text from pdf's.
Sample Code:
import java.io.*;
import org.apache.pdfbox.pdmodel.*;
import org.apache.pdfbox.util.*;
public class LittleExample {
public static void main(String[] args){
PDDocument pd;
BufferedWriter wr;
try {
// this is your pdf from which you would like to extract the text
File input = new File("/home/ottp/pdffiles/1.pdf");
// this is the target file to store the extracted text
File output = new File("/home/ottp/pdffiles/extracts/1.txt");
pd = PDDocument.load(input);
System.out.println(pd.getNumberOfPages());
System.out.println(pd.isEncrypted());
pd.save("CopyOfInvoice.pdf")
PDFTextStripper stripper = new PDFTextStripper();
wr = new BufferedWriter(new OutputStreamWriter(new FileOutputStream(output)));
stripper.writeText(pd, wr);
if (pd != null) {
pd.close();
}
// close and flush the output stream
wr.close();
} catch (Exception e){
e.printStackTrace();
}
}
}
I've written a simple Groovy script (below) to set the values of four of the ID3v1 and ID3v2 tag fields in mp3 files using the JAudioTagger library. The script successfully makes the changes but it also deletes the first 5 to 10 seconds of some of the files, other files are unaffected. It's not a big problem, but if anyone knows a simple fix, I would be grateful. All the files are from the same source, all have v1 and v2 tags, I can find no obvious difference in the source files to explain it.
import org.jaudiotagger.*
java.util.logging.Logger.getLogger("org.jaudiotagger").setLevel(java.util.logging.Level.OFF)
Integer trackNum = 0
Integer totalFiles = 0
Integer invalidFiles = 0
validMP3File = true
def dir = new File(/D:\Users\Jeremy\Music\Speech Radio\Unlistened\Z Temp Files to MP3 Tagged/)
dir.eachFile({curFile ->
totalFiles ++
try {
mp3File = org.jaudiotagger.audio.AudioFileIO.read(curFile)
} catch (org.jaudiotagger.audio.exceptions.CannotReadException e) {
validMP3File = false
invalidFiles ++
}
// Get the file name excluding the extension
baseFilename = org.jaudiotagger.audio.AudioFile.getBaseFilename(curFile)
// Check that it is an MP3 file
if (validMP3File) {
if (mp3File.getAudioHeader().getEncodingType() != 'mp3') {
validMP3File = false
invalidFiles ++
}
}
if (validMP3File) {
trackNum ++
if (mp3File.hasID3v1Tag()) {
curTagv1 = mp3File.getID3v1Tag()
} else {
curTagv1 = new org.jaudiotagger.tag.id3.ID3v1Tag()
}
if (mp3File.hasID3v2Tag()) {
curTagv2 = mp3File.getID3v2TagAsv24()
} else {
curTagv2 = new org.jaudiotagger.tag.id3.ID3v23Tag()
}
curTagv1.setField(org.jaudiotagger.tag.FieldKey.TITLE, baseFilename)
curTagv2.setField(org.jaudiotagger.tag.FieldKey.TITLE, baseFilename)
curTagv1.setField(org.jaudiotagger.tag.FieldKey.ARTIST, "BBC Radio")
curTagv2.setField(org.jaudiotagger.tag.FieldKey.ARTIST, "BBC Radio")
curTagv1.setField(org.jaudiotagger.tag.FieldKey.ALBUM, "BBC Radio - 20130205")
curTagv2.setField(org.jaudiotagger.tag.FieldKey.ALBUM, "BBC Radio - 20130205")
curTagv1.setField(org.jaudiotagger.tag.FieldKey.TRACK, trackNum.toString())
curTagv2.setField(org.jaudiotagger.tag.FieldKey.TRACK, trackNum.toString())
mp3File.setID3v1Tag(curTagv1)
mp3File.setID3v2Tag(curTagv2)
mp3File.save()
}
})
println """$trackNum tracks created from $totalFiles files with $invalidFiles invalid files"""
I'm still investigating and it appears that there is no problem with JAudioTagger. Before setting the tags, I use Total Recorder to reduce the quality of the download from 128kbps, 44,100Hz to 56kbps, 22,050Hz. This reduces the file size to less than half and the quality is fine for speech radio.
If I run my script on the original files, none of the audio track is deleted. The deletion of the first part of the audio track only occurs with the files that have been processed by Total Recorder.
Looking at the JAudioTagger logging for these files, there does appear to be a problem with the header:
Checking further because the ID3 Tag ends at 0x23f9 but the mp3 audio doesnt start until 0x7a77
Confirmed audio starts at 0x7a77 whether searching from start or from end of ID3 tag
This check is not performed for files that have not been processed by Total Recorder.
The log of the header read operation also shows (for a 27 minute track):
trackLength:06:52
It looks as though I shall have to find a new MP3 file editor!
Instead of
mp3File.save()
could you try:
mp3File.commit()
No idea if it will help, but that seems to be the documented method?