From 042c5cc16f4a41efd538b89e7d720f4b93781824 Mon Sep 17 00:00:00 2001
From: "Matthew D. Scholefield" <matthew331199@gmail.com>
Date: Mon, 20 Apr 2020 09:21:43 -0500
Subject: [PATCH] Add train_speech_commands.sh This is a script runs the the
 training process for a given word in the speech commands dataset

---
 train_speech_commands.sh | 87 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 87 insertions(+)
 create mode 100755 train_speech_commands.sh

diff --git a/train_speech_commands.sh b/train_speech_commands.sh
new file mode 100755
index 00000000..fcffff69
--- /dev/null
+++ b/train_speech_commands.sh
@@ -0,0 +1,87 @@
+#!/usr/bin/env bash
+
+cd "$(dirname "$0")"  # Cd to script location
+set -eE
+
+dataset_url="http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz"
+dataset_md5="6b74f3901214cb2c2934e98196829835"
+dataset_filename="speech_commands_v0.02.tar.gz"
+dataset_folder="speech_commands"
+demo_class="marvin"
+
+mkdir -p "data"
+echo "Downloading speech commands dataset..."
+cur_md5=$(md5sum "data/$dataset_filename" | awk '{print $1}')
+if [ "$cur_md5" != "$dataset_md5" ] || [ ! -d "data/$dataset_folder" ]; then
+    if [ "$cur_md5" != "$dataset_md5" ]; then
+        wget "$dataset_url" -O "data/$dataset_filename"
+    fi
+    pushd "data/$dataset_folder"
+    tar xvf "../$dataset_filename"
+    popd 2>/dev/null
+else
+    echo "Skipping, already downloaded."
+fi
+
+pushd "data/$dataset_folder"
+classes=$(find * -maxdepth 0 -type d -name '[a-zA-Z]*')
+for class in $classes; do
+    if [ -f "tags-$class.txt" ]; then
+        echo "Already generated tags-$class.txt."
+        continue
+    fi
+    echo "Generating tags-$class.txt..."
+    {
+        find "$class" -name '*.wav' | {
+            while read line; do
+                printf "${line%.wav}\twake-word\n"
+            done
+        }
+        for other_class in $classes; do
+            if [ "$class" = "$other_class" ]; then
+                continue
+            fi
+            find "$other_class" -name '*.wav' | {
+                while read line; do
+                    printf "${line%.wav}\tnot-wake-word\n"
+                done
+            }
+        done
+    } > "tags-$class.txt"
+done
+popd 2>/dev/null
+
+./setup.sh
+source .venv/bin/activate
+
+echo "Dataset import complete."
+
+mkdir -p models/
+train_command="precise-train models/$demo_class.net data/$dataset_folder --tags-file data/$dataset_folder/tags-$demo_class.txt -e 3 --batch-size 128 --sensitivity 0.5"
+echo ""
+echo "Training $demo_class model with command:"
+echo "$ $train_command"
+echo ""
+echo "Press any key to begin model training..."
+read -n 1
+eval "$train_command"
+
+echo "Model saved to models/$demo_class.net"
+echo ""
+
+test_command="precise-test models/$demo_class.net data/$dataset_folder --tags-file data/$dataset_folder/tags-$demo_class.txt"
+echo "Testing $demo_class model with command:"
+echo "$ $test_command"
+echo ""
+echo "Press any key to test model..."
+read -n 1
+eval "$test_command"
+
+listen_command="precise-listen models/$demo_class.net --sensitivity 0.5"
+echo "Running $demo_class model against microphone with command:"
+echo "$ $listen_command"
+echo ""
+echo "Note: This will continuously listen to the microphone and should activate with the word \"$demo_class\". When you are done testing you can exit with Ctrl+C."
+echo "Press any key to test against microphone..."
+read -n 1
+eval "$listen_command"