From 00048efad2f80091dc7ef2f340b3ed73141ed892 Mon Sep 17 00:00:00 2001
From: Florian Quirin <info@bytemind.de>
Date: Thu, 19 Jan 2023 20:41:46 +0100
Subject: [PATCH] added sherpa-ncnn

---
 sherpa-ncnn/README.md   | 11 +++++++++++
 sherpa-ncnn/install.sh  | 21 +++++++++++++++++++++
 sherpa-ncnn/run-test.sh | 35 +++++++++++++++++++++++++++++++++++
 whisper-cpp/run-test.sh |  4 +++-
 4 files changed, 70 insertions(+), 1 deletion(-)
 create mode 100644 sherpa-ncnn/README.md
 create mode 100644 sherpa-ncnn/install.sh
 create mode 100644 sherpa-ncnn/run-test.sh

diff --git a/sherpa-ncnn/README.md b/sherpa-ncnn/README.md
new file mode 100644
index 0000000..137fbb7
--- /dev/null
+++ b/sherpa-ncnn/README.md
@@ -0,0 +1,11 @@
+# Next-gen Kaldi - Sherpa ncnn
+
+**Sherpa**: Real-time (online/streaming) speech recognition using next-gen Kaldi with ncnn.  
+**ncnn**: High-performance neural network inference framework optimized for the mobile platform  
+  
+Repositories:
+- K2 Sherpa ncnn: https://github.com/k2-fsa/sherpa-ncnn
+- ncnn: https://github.com/tencent/ncnn
+
+Tested with:
+- Arm64 - Debian 11
diff --git a/sherpa-ncnn/install.sh b/sherpa-ncnn/install.sh
new file mode 100644
index 0000000..688e5c9
--- /dev/null
+++ b/sherpa-ncnn/install.sh
@@ -0,0 +1,21 @@
+#!/bin/bash
+echo "Installing Sherpa-ncnn ..."
+sudo apt update
+sudo apt install -y --no-install-recommends git build-essential
+echo "Cloning code ..."
+git clone https://github.com/k2-fsa/sherpa-ncnn
+echo "Running build process ..."
+cd sherpa-ncnn
+mkdir build
+cd build
+cmake -DCMAKE_BUILD_TYPE=Release ..
+make -j3
+cd ..
+echo "Downloading models ..."
+mkdir -p models
+cd models
+if [ ! -d "sherpa-ncnn-conv-emformer-transducer-small-2023-01-09/" ]; then
+	wget https://github.com/fquirin/speech-recognition-experiments/releases/download/v1.0.0/sherpa-ncnn-conv-emformer-transducer-small-2023-01-09.zip
+	unzip sherpa-ncnn-conv-emformer-transducer-small-2023-01-09.zip -d sherpa-ncnn-conv-emformer-transducer-small-2023-01-09
+fi
+echo "DONE"
diff --git a/sherpa-ncnn/run-test.sh b/sherpa-ncnn/run-test.sh
new file mode 100644
index 0000000..fe70bd0
--- /dev/null
+++ b/sherpa-ncnn/run-test.sh
@@ -0,0 +1,35 @@
+#!/usr/bin/env bash
+set -e
+
+cd sherpa-ncnn
+EXE="./build/bin/sherpa-ncnn"
+which $EXE
+
+MODEL="./models/sherpa-ncnn-conv-emformer-transducer-small-2023-01-09"
+THREADS=2
+echo "Model: $MODEL"
+echo "Threads: $THREADS"
+echo ""
+
+waves=(
+"../../test-files/en_speech_jfk_11s.wav"
+"../../test-files/en_sh_lights_70pct_4s.wav"
+)
+
+start_t=$(date +"%s.%N")
+for wave in ${waves[@]}; do
+  echo "Transcribing file: $wave"
+  time $EXE \
+    $MODEL/tokens.txt \
+    $MODEL/encoder_jit_trace-pnnx.ncnn.param \
+    $MODEL/encoder_jit_trace-pnnx.ncnn.bin \
+    $MODEL/decoder_jit_trace-pnnx.ncnn.param \
+    $MODEL/decoder_jit_trace-pnnx.ncnn.bin \
+    $MODEL/joiner_jit_trace-pnnx.ncnn.param \
+    $MODEL/joiner_jit_trace-pnnx.ncnn.bin \
+    $wave "$THREADS"
+    #$wave 4 greedy_search
+  echo ""
+done
+finish_t=$(date +"%s.%N")
+echo "Total: $(expr $finish_t-$start_t | bc)s"
diff --git a/whisper-cpp/run-test.sh b/whisper-cpp/run-test.sh
index 0f40644..aa656e2 100644
--- a/whisper-cpp/run-test.sh
+++ b/whisper-cpp/run-test.sh
@@ -13,7 +13,9 @@ echo ""
 
 start_t=$(date +"%s.%N")
 for wave in ${waves[@]}; do
+  echo "Transcribing file: $wave"
   ./main -m "models/ggml-${MODEL}.bin" -f "$wave" -t $THREADS
+  echo ""
 done
 finish_t=$(date +"%s.%N")
-echo "Took: $(expr $finish_t-$start_t | bc)s"
+echo "Total: $(expr $finish_t-$start_t | bc)s"