clutch of new models,

changes to make convert specs work
some efficiencies in the model code

js/BirdNet2.4.js

@@ -91,8 +91,7 @@ onmessage = async (e) => {
                         return myModel.makeSpectrogram(bufferTensor);
                     });
                     let spec = myModel.fixUpSpecBatch(tf.expandDims(imageTensor, 0), spec_height, spec_width);
-                    const spec_max = tf.max(spec);
-                    return spec.mul(255).div(spec_max).dataSync();
+                    return spec.dataSync();
                 });
 
                 response = {
@@ -197,13 +196,26 @@ class Model {
             */
             //specBatch = tf.log1p(specBatch).mul(20);
             // Swap axes to fit output shape
+
             specBatch = tf.transpose(specBatch, [0, 2, 1]);
             specBatch = tf.reverse(specBatch, [1]);
             // Add channel axis
-            specBatch = tf.expandDims(specBatch, -1);
+
             //specBatch = tf.slice4d(specBatch, [0, 1, 0, 0], [-1, img_height, img_width, -1]);
-            specBatch = tf.image.resizeBilinear(specBatch, [img_height, img_width], true);
-            return  this.normalise(specBatch)
+            //specBatch = tf.image.resizeBilinear(specBatch, [img_height, img_width], true);
+            // Slice to exclude the bottom 10 rows
+            const sliced_tensor = tf.slice3d(specBatch, [0, 1, 0], [-1, img_height - 10, img_width])
+
+            // Slice the bottom 10 rows
+            const bottomRows = tf.slice3d(sliced_tensor, [0, img_height - 20, 0], [-1, 10, img_width]);
+
+            // Reduce the values of the bottom rows
+            const attenuatedRows = bottomRows.div(tf.scalar(5));
+
+            // Concatenate the sliced tensor with the bottom rows
+            specBatch = tf.concat([sliced_tensor, attenuatedRows], 1)
+            specBatch = this.normalise(specBatch)
+            return  tf.expandDims(specBatch, -1);
         })
     }
     normalise(spec) {

js/model.js

@@ -120,8 +120,7 @@ onmessage = async (e) => {
                         return myModel.makeSpectrogram(bufferTensor);
                     });
                     let spec = myModel.fixUpSpecBatch(tf.expandDims(imageTensor, 0), spec_height, spec_width);
-                    const spec_max = tf.max(spec);
-                    return spec.mul(255).div(spec_max).dataSync();
+                    return spec.dataSync();
                 });
                 response = {
                     message: "spectrogram",
@@ -365,13 +364,26 @@ class Model {
             */
             //specBatch = tf.log1p(specBatch).mul(20);
             // Swap axes to fit output shape
+
             specBatch = tf.transpose(specBatch, [0, 2, 1]);
             specBatch = tf.reverse(specBatch, [1]);
             // Add channel axis
-            specBatch = tf.expandDims(specBatch, -1);
+
             //specBatch = tf.slice4d(specBatch, [0, 1, 0, 0], [-1, img_height, img_width, -1]);
-            specBatch = tf.image.resizeBilinear(specBatch, [img_height, img_width], true);
-            return  this.normalise(specBatch)
+            //specBatch = tf.image.resizeBilinear(specBatch, [img_height, img_width], true);
+            // Slice to exclude the bottom 10 rows
+            const sliced_tensor = tf.slice3d(specBatch, [0, 1, 0], [-1, img_height - 10, img_width])
+
+            // Slice the bottom 10 rows
+            const bottomRows = tf.slice3d(sliced_tensor, [0, img_height - 20, 0], [-1, 10, img_width]);
+
+            // Reduce the values of the bottom rows
+            const attenuatedRows = bottomRows.div(tf.scalar(5));
+
+            // Concatenate the sliced tensor with the bottom rows
+            specBatch = tf.concat([sliced_tensor, attenuatedRows], 1)
+            specBatch = this.normalise(specBatch)
+            return  tf.expandDims(specBatch, -1);
         })
     }
     normalise_audio_batch = (tensor) => {
@@ -415,11 +427,8 @@ class Model {
         audioBuffer.dispose();
         const bufferList = this.version !== 'v4' ? this.normalise_audio_batch(buffers) : buffers;
         const specBatch = tf.tidy(() => {
-            const bufferArray = tf.unstack(bufferList);
-            const toStack = bufferArray.map(x => {
-                return this.makeSpectrogram(x)
-            })
-            return this.fixUpSpecBatch(tf.stack(toStack))
+            const toStack = tf.unstack(bufferList).map(x => this.makeSpectrogram(x));
+            return this.fixUpSpecBatch(tf.stack(toStack));
         });
         buffers.dispose();
         bufferList.dispose();

js/ui.js

@@ -2791,7 +2791,7 @@ function centreSpec(){
             }
         },
         D: function (e) {
-            if (( e.ctrlKey || e.metaKey)) worker.postMessage({ action: 'create-dataset' });
+            if (( e.ctrlKey || e.metaKey)) worker.postMessage({ action: 'convert-dataset' });
         },
         e: function (e) {
             if (( e.ctrlKey || e.metaKey) && region) exportAudio();
@@ -3105,7 +3105,7 @@ function centreSpec(){
         metadata = undefined
     }) {
         fileLoaded = true, locationID = location;
-        bufferBegin = fileBegin;
+
         clearTimeout(loadingTimeout)
         // Clear the loading animation
         DOM.loading.classList.add('d-none');

js/worker.js

@@ -8,6 +8,9 @@ const ffmpeg = require('fluent-ffmpeg');
 const png = require('fast-png');
 const {utimesSync} = require('utimes');
 const {writeToPath} = require('@fast-csv/format');
+
+const { execSync } = require('child_process');
+
 const merge = require('lodash.merge');
 import { State } from './state.js';
 import { sqlite3 } from './database.js';
@@ -16,9 +19,9 @@ import {extractWaveMetadata} from './metadata.js';
 let isWin32 = false;
 
 const DATASET = true;
+const DATABASE = 'archive_test'
 const adding_chirpity_additions = false;
-const dataset_database = DATASET;
-const DATASET_SAVE_LOCATION = "/media/matt/36A5CC3B5FA24585/DATASETS/MISSING_PNGS";
+const DATASET_SAVE_LOCATION = "/media/matt/36A5CC3B5FA24585/DATASETS/ATTENUATED_pngs/call";
 let ntsuspend;
 if (process.platform === 'win32') {
     ntsuspend = require('ntsuspend');
@@ -194,7 +197,7 @@ const createDB = async (file) => {
 
     if (archiveMode) {
         for (let i = 0; i < LABELS.length; i++) {
-            const [sname, cname] = LABELS[i].replaceAll("'", "''").split('_');
+            const [sname, cname] = LABELS[i].split('_');
             await db.runAsync('INSERT INTO species VALUES (?,?,?)', i, sname, cname);
         }
     } else {
@@ -242,7 +245,7 @@ async function loadDB(path) {
     modelLabels.push("Unknown Sp._Unknown Sp.");
     const num_labels = modelLabels.length;
     LABELS = modelLabels; // these are the default english labels
-    const file = dataset_database ? p.join(path, `archive_dataset${num_labels}.sqlite`) : p.join(path, `archive${num_labels}.sqlite`)
+    const file = DATASET ? p.join(path, `${DATABASE}${num_labels}.sqlite`) : p.join(path, `archive${num_labels}.sqlite`)
 
     if (!fs.existsSync(file)) {
         await createDB(file);
@@ -1257,41 +1260,37 @@ function checkBacklog(ffmpegCommand = null) {
             console.warn('Ffmpeg process already exited in check backlog call')
             return 
         }
-        let firstRun = true, hysteresis_factor = 2;
+        const hysteresis_factor = 2;
 
         if (!aborted && AUDIO_BACKLOG < NUM_WORKERS * hysteresis_factor && ! STATE.processingPaused[pid]) {
+            console.log('no need to check further, backlog: ', AUDIO_BACKLOG, ' pid paused: ', pid, STATE.processingPaused[pid])
             resolve(true);
             return;
         }
-
         const interval = setInterval(() => {
             const backlog = AUDIO_BACKLOG;
+            console.log('Backlog in interval is:', backlog)
             if (aborted) {
+                console.log('clearing interval, aborted')
                 clearInterval(interval);
                 resolve(false);
                 return;
             }
-
-            if (firstRun) {
-                DEBUG && console.log('backlog (initial):', backlog);
-                firstRun = false; // Ensure this logs only on the first call
-            }
-
-            if (backlog >= NUM_WORKERS * hysteresis_factor) {
-                if (! STATE.processingPaused[pid]) {
+            if (! STATE.processingPaused[pid] && backlog >= NUM_WORKERS * hysteresis_factor) {
                     pauseFfmpeg(ffmpegCommand, pid);
                     DEBUG && console.log(`pause signal sent to ffmpeg(${pid}), backlog: ${backlog}`);
                     STATE.processingPaused[pid] = true;
-                }
-            } else if (backlog <= NUM_WORKERS) {
-                DEBUG && console.log('backlog (final):', backlog);
-                if (STATE.processingPaused[pid]) {
-                     resumeFfmpeg(ffmpegCommand, pid);
-                     DEBUG && console.log(`resume signal sent to ffmpeg(${pid}), backlog: ${backlog}`);
-                     STATE.processingPaused[pid] = false;
-                }
-                clearInterval(interval); // Backlog is within limits, stop checking
-                resolve(true); // Resolve the promise
+            }
+            else if (backlog <= NUM_WORKERS) {
+            DEBUG && console.log('backlog (final):', backlog);
+            if (STATE.processingPaused[pid]) {
+                resumeFfmpeg(ffmpegCommand, pid);
+                DEBUG && console.log(`resume signal sent to ffmpeg(${pid}), backlog: ${backlog}`);
+                STATE.processingPaused[pid] = false;
+            }
+            console.log('clearing interval, pressure eased')
+            clearInterval(interval); // Backlog is within limits, stop checking
+            resolve(true); // Resolve the promise
             }
         }, 200);
     });
@@ -1304,6 +1303,7 @@ function pauseFfmpeg(ffmpegCommand, pid){
         DEBUG && console.log(message)
     } else {
         ffmpegCommand.kill('SIGSTOP')
+        console.log('paused ', pid)
     }
 }
 
@@ -1314,6 +1314,7 @@ function resumeFfmpeg(ffmpegCommand, pid){
         DEBUG && console.log(message)
     } else {
         ffmpegCommand.kill('SIGCONT')
+        console.log('resumded ', pid)
     }
 }
 
@@ -1358,6 +1359,8 @@ const getPredictBuffers = async ({
 }
 
 async function processAudio (file, start, end, chunkStart, highWaterMark, samplesInBatch){
+    const MAX_CHUNKS = NUM_WORKERS * 2;
+    let isPaused = false;
     return new Promise((resolve, reject) => {
         // Many compressed files start with a small section of silence due to encoder padding, which affects predictions
         // To compensate, we move the start back a small amount, and slice the data to remove the silence
@@ -1378,12 +1381,27 @@ async function processAudio (file, start, end, chunkStart, highWaterMark, sample
                 reject(error)
             }
         });
-        command.on('progress', (progress) => {
-            DEBUG && console.log('progress: ', progress.timemark)
-            checkBacklog(command)
-        })
+        // command.on('progress', (progress) => {
+        //     DEBUG && console.log('progress: ', progress.timemark)
+        //     checkBacklog(command)
+        // })
         const STREAM = command.pipe();
         STREAM.on('data', (chunk) => {
+            if (! isPaused && AUDIO_BACKLOG >= MAX_CHUNKS){
+                isPaused = true;
+                const pid = command.ffmpegProc?.pid
+                pid && pauseFfmpeg(command, pid)
+                console.log('about to set the interval ', pid)
+                const interval = setInterval(() =>{
+                    console.log('Backlog', AUDIO_BACKLOG)
+                    if (AUDIO_BACKLOG < NUM_WORKERS){
+                        resumeFfmpeg(command, pid)
+                        console.log('resumed ', pid)
+                        isPaused = false;
+                        clearInterval(interval)
+                    }
+                }, 100)
+            }
             if (aborted) {
                 STREAM.destroy();
                 return;
@@ -1614,29 +1632,86 @@ const speciesMatch = (path, sname) => {
     return species.includes(sname)
 }
 
+function findFile(pathParts, filename, species) {
+    const baseDir = pathParts.slice(0, 5).concat(['XC_ALL_mp3']).join(p.sep);
+
+    // List of suffixes to check, in order
+    const suffixes = ['', ' (call)', ' (fc)', ' (nfc)', ' (song)'];
+
+    // Extract existing suffix from species, if present
+    const suffixPattern = / \((call|fc|nfc|song)\)$/;
+    let speciesBase = species;
+    let existingSuffix = '';
+
+    const match = species.match(suffixPattern);
+    if (match) {
+        existingSuffix = match[0]; // e.g., " (call)"
+        speciesBase = species.replace(suffixPattern, ''); // Remove suffix
+    }
+
+    // First, check the species with its existing suffix
+    if (existingSuffix) {
+        const folder = p.join(baseDir, species);
+        const filePath = p.join(folder, filename + '.mp3');
+        if (fs.existsSync(filePath)) {
+            console.log(`File found: ${filePath}`);
+            return [filePath, species];
+        }
+    }
+
+    // Check species with other suffixes, removing the existing one
+    for (const suffix of suffixes) {
+        if (suffix === existingSuffix) continue; // Skip the suffix already checked
+        const found_calltype = speciesBase + suffix
+        const folder = p.join(baseDir, found_calltype);
+        const filePath = p.join(folder, filename + '.mp3');
+        if (fs.existsSync(filePath)) {
+            console.log(`File found: ${filePath}`);
+
+            return [filePath, found_calltype];
+        }
+    }
+
+    console.log('File not found in any directory');
+    return [null, null];
+}
 const convertSpecsFromExistingSpecs = async (path) => {
-    path ??= '/mnt/608E21D98E21A88C/Users/simpo/PycharmProjects/Data/New_Dataset';
+    path ??= '/media/matt/36A5CC3B5FA24585/DATASETS/MISSING/NEW_DATASET_WITHOUT_ALSO_MERGED';
     const file_list = await getFiles([path], true);
     for (let i = 0; i < file_list.length; i++) {
+        if (i % 100 === 0){
+            console.log(`${i} records processed`)
+        }
         const parts = p.parse(file_list[i]);
-        let species = parts.dir.split(p.sep);
-        species = species[species.length - 1];
+        let path_parts = parts.dir.split(p.sep);
+        let species = path_parts[path_parts.length - 1];
+        const species_parts = species.split('~');
+        species = species_parts[1] + '~' + species_parts[0]
         const [filename, time] = parts.name.split('_');
         const [start, end] = time.split('-');
-        const path_to_save = path.replace('New_Dataset', 'New_Dataset_Converted') + p.sep + species;
-        const file_to_save = p.join(path_to_save, parts.base);
+        // const path_to_save = path.replace('New_Dataset', 'New_Dataset_Converted') + p.sep + species;
+        let path_to_save = '/media/matt/36A5CC3B5FA24585/DATASETS/ATTENUATED_pngs/converted' + p.sep + species;
+        let file_to_save = p.join(path_to_save, parts.base);
         if (fs.existsSync(file_to_save)) {
             DEBUG && console.log("skipping file as it is already saved")
         } else {
-            const file_to_analyse = parts.dir.replace('New_Dataset', 'XC_ALL_mp3') + p.sep + filename + '.mp3';
-            const AudioBuffer = await fetchAudioBuffer({
+            const [file_to_analyse, confirmed_species_folder] = findFile(path_parts, filename, species)
+            path_to_save = path_to_save.replace(species, confirmed_species_folder)
+            file_to_save = p.join(path_to_save, parts.base);
+            if (fs.existsSync(file_to_save)){
+                console.log("skipping file as it is already saved")
+                continue;
+            }
+            if (! file_to_analyse) continue
+            //parts.dir.replace('MISSING/NEW_DATASET_WITHOUT_ALSO_MERGED', 'XC_ALL_mp3') + p.sep + filename + '.mp3';
+            const [AudioBuffer, begin] = await fetchAudioBuffer({
                 start: parseFloat(start), end: parseFloat(end), file: file_to_analyse
             })
             if (AudioBuffer) {  // condition to prevent barfing when audio snippet is v short i.e. fetchAudioBUffer false when < 0.1s
                 if (++workerInstance === NUM_WORKERS) {
                     workerInstance = 0;
                 }
-                const buffer = AudioBuffer.getChannelData(0);
+                const buffer = getMonoChannelData(AudioBuffer);
                 predictWorkers[workerInstance].postMessage({
                     message: 'get-spectrogram',
                     filepath: path_to_save,
@@ -3221,7 +3296,7 @@ async function onChartRequest(args) {
     const pointStart = dateRange.start ??= Date.UTC(2020, 0, 0, 0, 0, 0);
     UI.postMessage({
         event: 'chart-data', // Restore species name
-        species: args.species ? args.species.replace("''", "'") : undefined,
+        species: args.species ? args.species : undefined,
         results: results,
         rate: rate,
         total: total,

labels/V2.4/BirdNET_GLOBAL_6K_V2.4_Labels_en.txt

@@ -1722,7 +1722,7 @@ Cuculus optatus_Oriental Cuckoo
 Cuculus poliocephalus_Lesser Cuckoo
 Cuculus rochii_Madagascar Cuckoo
 Cuculus saturatus_Himalayan Cuckoo
-Cuculus solitarius_Red-chested Cuckoo
+Cuculus solitarius_Red-chestiberiaeed Cuckoo
 Culicicapa ceylonensis_Gray-headed Canary-Flycatcher
 Culicicapa helianthea_Citrine Canary-Flycatcher
 Culicivora caudacuta_Sharp-tailed Tyrant