diff --git a/requirements.txt b/requirements.txt
index 074426f35..8034f72a8 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -17,3 +17,4 @@ matplotlib
 pillow
 pyarrow==20.0.0
 soundfile
+resampy
diff --git a/scripts/advanced.php b/scripts/advanced.php
index 2a65d1b61..9b4096cd5 100644
--- a/scripts/advanced.php
+++ b/scripts/advanced.php
@@ -329,7 +329,7 @@
       </td></tr><tr><td>
       If different than 0 (keep all), defines the number of files to keep for each species, with priority given to files with higher confidence. This value does not include files from the last 7 days, these new files are protected against auto-deletion.
       </td></tr><tr><td>
-      Note only the spectrogram and audio files are deleted, the obsevation data remains in the database.
+      Note only the spectrogram and audio files are deleted, the observation data remains in the database.
       The files protected through the "lock" icon are also not affected.
       <br>
       <button type="submit" name="run_species_count" value="1" onclick="{this.innerHTML = 'Loading ... please wait.';this.classList.add('disabled')}"><i>[Click here for disk usage summary]</i></button>
@@ -345,7 +345,7 @@
       Set Channels to the number of channels supported by your sound card. 32 max.<br><br>
       <label for="recording_length">Recording Length: </label>
       <input name="recording_length" oninput="document.getElementsByName('extraction_length')[0].setAttribute('max', this.value);" type="number" style="width:3em;" min="3" max="60" step="1" value="<?php print($newconfig['RECORDING_LENGTH']);?>" required/><br>
-      Set Recording Length in seconds between 6 and 60. Multiples of 3 are recommended, as BirdNET analyzes in 3-second chunks.<br><br> 
+      Set Recording Length in seconds between 6 and 60. Multiples of 3 are recommended for BirdNET models, multiples of 5 for Perch.<br><br>
       <label for="extraction_length">Extraction Length: </label>
       <input name="extraction_length" oninput="this.setAttribute('max', document.getElementsByName('recording_length')[0].value);" type="number" style="width:3em;" min="3" value="<?php print($newconfig['EXTRACTION_LENGTH']);?>" /><br>
       Set Extraction Length to something less than your Recording Length. Min=3 Max=Recording Length<br><br>
diff --git a/scripts/birdnet_recording.sh b/scripts/birdnet_recording.sh
index c714a16a6..839c38c50 100755
--- a/scripts/birdnet_recording.sh
+++ b/scripts/birdnet_recording.sh
@@ -4,24 +4,32 @@ source /etc/birdnet/birdnet.conf
 
 loop_ffmpeg(){
   while true;do
-    if ! ffmpeg -hide_banner -xerror -loglevel $LOGGING_LEVEL -nostdin ${1} -i ${2} -vn -map a:0 -acodec pcm_s16le -ac 2 -ar 48000 -f segment -segment_format wav -segment_time ${RECORDING_LENGTH} -strftime 1 ${RECS_DIR}/StreamData/%F-birdnet-RTSP_${3}-%H:%M:%S.wav
+    if ! ffmpeg -hide_banner -xerror -loglevel $LOGGING_LEVEL -nostdin ${1} -i ${2} -vn -map a:0 -acodec pcm_s16le -ac 1 -ar ${SAMPLERATE} -f segment -segment_format wav -segment_time ${RECORDING_LENGTH} -strftime 1 ${RECS_DIR}/StreamData/%F-birdnet-RTSP_${3}-%H:%M:%S.wav
     then
       sleep 1
     fi
   done
 }
 
-# Read the logging level from the configuration option
-LOGGING_LEVEL="${LogLevel_BirdnetRecordingService}"
-# If empty for some reason default to log level of error
-[ -z $LOGGING_LEVEL ] && LOGGING_LEVEL='error'
+# Set logging level, default to 'error' if not set
+LOGGING_LEVEL="${LogLevel_BirdnetRecordingService:-error}"
+
 # Additionally if we're at debug or info level then allow printing of script commands and variables
 if [ "$LOGGING_LEVEL" == "info" ] || [ "$LOGGING_LEVEL" == "debug" ];then
   # Enable printing of commands/variables etc to terminal for debugging
   set -x
 fi
 
-[ -z $RECORDING_LENGTH ] && RECORDING_LENGTH=15
+REC_CARD="${REC_CARD:-default}"
+RECORDING_LENGTH="${RECORDING_LENGTH:-15}"
+
+# Set sample rate based on the model
+if [[ "$MODEL" == "Perch_v2" ]]; then
+  SAMPLERATE=32000
+else
+  SAMPLERATE=48000
+fi
+
 [ -d $RECS_DIR/StreamData ] || mkdir -p $RECS_DIR/StreamData
 
 if [ -n "${RTSP_STREAM}" ];then
@@ -50,12 +58,7 @@ else
   if pgrep arecord &> /dev/null ;then
     echo "Recording"
   else
-    if [ -z ${REC_CARD} ];then
-      arecord -f S16_LE -c${CHANNELS} -r48000 -t wav --max-file-time ${RECORDING_LENGTH}\
-	      	      	       --use-strftime ${RECS_DIR}/StreamData/%F-birdnet-%H:%M:%S.wav
-    else
-      arecord -f S16_LE -c${CHANNELS} -r48000 -t wav --max-file-time ${RECORDING_LENGTH}\
-        -D "${REC_CARD}" --use-strftime ${RECS_DIR}/StreamData/%F-birdnet-%H:%M:%S.wav
-    fi
+    arecord -f S16_LE -c${CHANNELS} -r ${SAMPLERATE} -t wav --max-file-time ${RECORDING_LENGTH}\
+          -D "${REC_CARD}" --use-strftime ${RECS_DIR}/StreamData/%F-birdnet-%H:%M:%S.wav
   fi
 fi
diff --git a/scripts/config.php b/scripts/config.php
index ae12ebc00..8081961a8 100644
--- a/scripts/config.php
+++ b/scripts/config.php
@@ -235,7 +235,7 @@ function() {
 <script>
   document.addEventListener('DOMContentLoaded', function() {
   document.getElementById('modelsel').addEventListener('change', function() {
-    if(this.value == "BirdNET_GLOBAL_6K_V2.4_Model_FP16"){ 
+    if(["BirdNET_GLOBAL_6K_V2.4_Model_FP16", "BirdNET-Go_classifier_20250916"].indexOf(this.value) > -1){
       document.getElementById("soft").style.display="unset";
     } else {
       document.getElementById("soft").style.display="none";
@@ -267,7 +267,7 @@ function sendTestNotification(e) {
       <label for="model">Select a Model: </label>
       <select id="modelsel" name="model" class="testbtn">
       <?php
-      $models = array("BirdNET_GLOBAL_6K_V2.4_Model_FP16", "BirdNET_6K_GLOBAL_MODEL");
+      $models = array("BirdNET_GLOBAL_6K_V2.4_Model_FP16", "BirdNET_6K_GLOBAL_MODEL", "BirdNET-Go_classifier_20250916", "Perch_v2");
       foreach($models as $modelName){
           $isSelected = "";
           if($config['MODEL'] == $modelName){
@@ -279,7 +279,7 @@ function sendTestNotification(e) {
       ?>
       </select>
       <br>
-      <span <?php if($config['MODEL'] == "BirdNET_6K_GLOBAL_MODEL") { ?>style="display: none"<?php } ?> id="soft">
+      <span <?php if(!in_array($config['MODEL'], ["BirdNET_GLOBAL_6K_V2.4_Model_FP16", "BirdNET-Go_classifier_20250916"])) { ?>style="display: none"<?php } ?> id="soft">
       <input type="checkbox" name="data_model_version" <?php if($config['DATA_MODEL_VERSION'] == 2) { echo "checked"; };?> >
       <label for="data_model_version">Species range model V2.4 - V2</label>  [ <a target="_blank" href="https://github.com/kahst/BirdNET-Analyzer/discussions/234">Info here</a> ]<br>
       <label for="sf_thresh">Species Occurrence Frequency Threshold [0.0005, 0.99]: </label>
@@ -387,11 +387,20 @@ function runProcess() {
       <br>
       <dd id="ddnewline">This is the BirdNET-Analyzer model, the most advanced BirdNET model to date. Currently it  supports over 6,000 species worldwide, giving quite good species coverage for people in most of the world.</dd>
       <br>
+      <dt>BirdNET-Go_classifier_20250916</dt>
+      <br>
+      <dd id="ddnewline">Custom TensorFlow Lite format AI model classifiers for enhanced bird and wildlife identification. Extends the capabilities of the base BirdNET v2.4 model. <a target="_blank" href="https://github.com/tphakala/birdnet-go-classifiers">External model</a></dd>
+      <br>
+      <dt>Perch_v2</dt>
+      <br>
+      <dd id="ddnewline">Perch is a bioacoustics model trained to classify nearly 15,000 species. This model has relatively high system requirements: RPi5 or Intel Haswell level performance and 4GB of RAM. <a target="_blank" href="https://www.kaggle.com/models/google/bird-vocalization-classifier/tensorFlow2/perch_v2">External model</a></dd>
+      <br>
       <dt>BirdNET_6K_GLOBAL_MODEL (2020)</dt>
       <br>
       <dd id="ddnewline">This is the BirdNET-Lite model, with bird sound recognition for more than 6,000 species worldwide. This has generally worse performance than the newer models but is kept as a legacy option.</dd>
       <br>
-      <dt>[ In-depth technical write-up on the models <a target="_blank" href="https://github.com/mcguirepr89/BirdNET-Pi/wiki/BirdNET-Pi:-some-theory-on-classification-&-some-practical-hints">here</a> ]</dt>
+      <dt>[ External models are downloaded if not yet present, so make sure your internet connection is stable ]</dt>
+      <dt>[ In-depth technical write-up on the BirdNET models <a target="_blank" href="https://github.com/mcguirepr89/BirdNET-Pi/wiki/BirdNET-Pi:-some-theory-on-classification-&-some-practical-hints">here</a> ]</dt>
       </dl>
       </td></tr></table><br>
 
diff --git a/scripts/update_birdnet_snippets.sh b/scripts/update_birdnet_snippets.sh
index fec853fb5..0a5169717 100755
--- a/scripts/update_birdnet_snippets.sh
+++ b/scripts/update_birdnet_snippets.sh
@@ -167,6 +167,7 @@ fi
 
 ensure_python_package inotify inotify
 ensure_python_package soundfile soundfile
+ensure_python_package resampy resampy
 
 if ! which inotifywait &>/dev/null;then
   ensure_apt_updated
diff --git a/scripts/utils/analysis.py b/scripts/utils/analysis.py
index 37869c60b..8b42edff1 100644
--- a/scripts/utils/analysis.py
+++ b/scripts/utils/analysis.py
@@ -1,5 +1,6 @@
 import logging
 import os
+import re
 import time
 
 import librosa
@@ -101,9 +102,10 @@ def filter_humans(predictions):
 
     # mask for humans
     human_mask = [False] * len(predictions)
+    human = re.compile('(Human|Conversation|Speech|Child_|Female_|Male_|Yell)')
     for i, prediction in enumerate(predictions):
         for p in prediction[:human_cutoff]:
-            if 'Human' in p[0]:
+            if human.match(p[0]):
                 human_mask[i] = True
                 break
 
diff --git a/scripts/utils/helpers.py b/scripts/utils/helpers.py
index 9c192ca47..a5044682d 100644
--- a/scripts/utils/helpers.py
+++ b/scripts/utils/helpers.py
@@ -108,7 +108,7 @@ def get_model_labels(model=None):
 
 def set_label_file():
     lang = get_language()
-    labels = [f'{label}_{lang[label]}\n' for label in get_model_labels()]
+    labels = [f'{label}_{lang.get(label, label)}\n' for label in get_model_labels()]
     file_name = os.path.join(MODEL_PATH, 'labels.txt')
     if os.path.islink(file_name):
         os.remove(file_name)
diff --git a/scripts/utils/models.py b/scripts/utils/models.py
index 5569ef639..adeea8daa 100644
--- a/scripts/utils/models.py
+++ b/scripts/utils/models.py
@@ -2,8 +2,10 @@
 import math
 import operator
 import os
+import tarfile
 
 import numpy as np
+import requests
 
 from .helpers import get_settings, get_model_labels, MODEL_PATH
 
@@ -50,6 +52,29 @@ def get_meta_model(model=None, version=None):
         return MDataModel2(conf.getfloat('SF_THRESH'))
 
 
+def download_file(url, file_path):
+    tmp_file = f"{file_path}_tmp"
+    session = requests.Session()
+    response = session.get(url, stream=True)
+    response.raise_for_status()
+    block_size = 32768
+
+    log.info('Downloading: %s' % os.path.basename(file_path))
+    downloaded_size = 0
+    try:
+        with open(tmp_file, "wb") as outfile:
+            for data in response.iter_content(block_size):
+                downloaded_size += len(data)
+                log.info(f'Progress: {downloaded_size}')
+                outfile.write(data)
+    except requests.exceptions.HTTPError as e:
+        if os.path.exists(tmp_file):
+            os.unlink(tmp_file)
+        raise e
+
+    os.rename(tmp_file, file_path)
+
+
 class Basemodel:
     chunk_duration = None
     sample_rate = None
@@ -58,8 +83,9 @@ class Basemodel:
     _output_layer = 0
 
     def __init__(self):
-        model_path = os.path.join(MODEL_PATH, f'{self.model_name}.tflite')
-        self.interpreter = tflite.Interpreter(model_path)
+        self.model_path = os.path.join(MODEL_PATH, f'{self.model_name}.tflite')
+        self.ensure_model()
+        self.interpreter = tflite.Interpreter(self.model_path)
         self.interpreter.allocate_tensors()
         input_details = self.interpreter.get_input_details()
         output_details = self.interpreter.get_output_details()
@@ -82,6 +108,9 @@ def set_meta_data(self, lat, lon, week):
     def get_species_list(self):
         return []
 
+    def ensure_model(self):
+        pass
+
 
 class BirdNet(Basemodel):
     chunk_duration = 3
@@ -182,10 +211,29 @@ def predict(self, chunk):
         exp_x = np.exp(logits - np.max(logits))  # Stabilizing to prevent overflow
         return self.label(exp_x / np.sum(exp_x))
 
+    def ensure_model(self):
+        if os.path.exists(self.model_path):
+            return
+        base_url = 'https://github.com/Nachtzuster/BirdNET-Pi/releases/download/v0.11'
+        file = 'Perch_v2.tar.gz'
+        tmp_file = os.path.join(MODEL_PATH, file)
+        download_file(f'{base_url}/{file}', tmp_file)
+        log.info(f'Extracting {file}...')
+        with tarfile.open(tmp_file, "r:gz") as tar:
+            tar.extractall(MODEL_PATH)
+        os.unlink(tmp_file)
+
 
 class BirdNETGo20250916(BirdNetV2_4):
     model_name = 'BirdNET-Go_classifier_20250916'
 
+    def ensure_model(self):
+        if os.path.exists(self.model_path):
+            return
+        base_url = 'https://raw.githubusercontent.com/tphakala/birdnet-go-classifiers/refs/heads/main/20250916'
+        for file in ['BirdNET-Go_classifier_20250916_Labels.txt', 'BirdNET-Go_classifier_20250916.tflite']:
+            download_file(f'{base_url}/{file}', os.path.join(MODEL_PATH, file))
+
 
 class MDataModel:
     model_name = None
diff --git a/tests/test_analysis.py b/tests/test_analysis.py
index 78b38d849..fee1383cd 100644
--- a/tests/test_analysis.py
+++ b/tests/test_analysis.py
@@ -2,10 +2,10 @@
 import unittest
 from unittest.mock import patch
 
+from scripts.utils.analysis import filter_humans
 from scripts.utils.analysis import run_analysis
 from scripts.utils.classes import ParseFileName
 from tests.helpers import TESTDATA, Settings
-from scripts.utils.analysis import filter_humans
 
 
 class TestRunAnalysis(unittest.TestCase):
@@ -57,13 +57,15 @@ def test_filter_humans_no_human(self, mock_load_settings):
         # Input detections without humans
         detections = [
             [('Bird_A', 0.9), ('Bird_B', 0.8)],
-            [('Bird_C', 0.7), ('Bird_D', 0.6)]
+            [('Bird_C', 0.9), ('Pacarina schumanni', 0.8)],
+            [('Bird_F', 0.7), ('Bird_F', 0.6)]
         ]
 
         # Expected output
         expected = [
             [('Bird_A', 0.9), ('Bird_B', 0.8)],
-            [('Bird_C', 0.7), ('Bird_D', 0.6)]
+            [('Bird_C', 0.9), ('Pacarina schumanni', 0.8)],
+            [('Bird_F', 0.7), ('Bird_F', 0.6)]
         ]
 
         # Run filter_humans
@@ -194,6 +196,60 @@ def test_filter_humans_with_human_deep(self, mock_load_settings):
         # Assertions
         self.assertEqual(result, expected)
 
+    @patch('scripts.utils.helpers._load_settings')
+    def test_filter_humans_for_perch(self, mock_load_settings):
+        mock_load_settings.return_value = Settings.with_defaults()
+
+        # Input detections with "humans" from FSD50K
+        # Child_speech_and_kid_speaking
+        # Conversation
+        # Female_singing
+        # Female_speech_and_woman_speaking
+        # Human_voice
+        # Male_singing
+        # Male_speech_and_man_speaking
+        # Speech
+        # Speech_synthesizer
+        # Yell
+
+        # Human_group_actions ??
+        detections = [
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_A', 0.8), ('Child_speech_and_kid_speaking', 0.75)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Conversation', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Female_singing', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Female_speech_and_woman_speaking', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Human_group_actions', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Human_voice', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Male_singing', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Male_speech_and_man_speaking', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Speech', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Speech_synthesizer', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+            [('Bird_B', 0.9), ('Yell', 0.8)],
+            [('Bird_A', 0.9), ('Bird_B', 0.8)],
+        ]
+
+        # Expected output
+        expected = [
+                       [('Human_Human', 0.0)]
+                   ] * 23
+
+        # Run filter_humans
+        result = filter_humans(detections)
+
+        # Assertions
+        self.assertEqual(result, expected)
+
 
 if __name__ == '__main__':
     unittest.main()