From 4a7d49af9550725b5095fdcba3306a075b3ea32e Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Sun, 30 Apr 2023 19:12:49 +0300
Subject: [PATCH] examples : fix + refactor Levenshtein distance

---
 examples/command.wasm/emscripten.cpp | 25 -------------------------
 examples/command/command.cpp         | 25 -------------------------
 examples/common.cpp                  | 24 ++++++++++++++++++++++++
 examples/common.h                    |  2 ++
 4 files changed, 26 insertions(+), 50 deletions(-)

diff --git a/examples/command.wasm/emscripten.cpp b/examples/command.wasm/emscripten.cpp
index 1cfd063..e739656 100644
--- a/examples/command.wasm/emscripten.cpp
+++ b/examples/command.wasm/emscripten.cpp
@@ -28,31 +28,6 @@ std::string g_transcribed   = "";
 
 std::vector<float> g_pcmf32;
 
-// compute similarity between two strings using Levenshtein distance
-static float similarity(const std::string & s0, const std::string & s1) {
-    const size_t len0 = s0.size() + 1;
-    const size_t len1 = s1.size() + 1;
-
-    std::vector<int> col(len1, 0);
-    std::vector<int> prevCol(len1, 0);
-
-    for (size_t i = 0; i < len1; i++) {
-        prevCol[i] = i;
-    }
-
-    for (size_t i = 0; i < len0; i++) {
-        col[0] = i;
-        for (size_t j = 1; j < len1; j++) {
-            col[j] = std::min(std::min(1 + col[j - 1], 1 + prevCol[j]), prevCol[j - 1] + (s0[i - 1] == s1[j - 1] ? 0 : 1));
-        }
-        col.swap(prevCol);
-    }
-
-    const float dist = prevCol[len1 - 1];
-
-    return 1.0f - (dist / std::max(s0.size(), s1.size()));
-}
-
 void command_set_status(const std::string & status) {
     std::lock_guard<std::mutex> lock(g_mutex);
     g_status = status;
diff --git a/examples/command/command.cpp b/examples/command/command.cpp
index 2b9440a..54e3549 100644
--- a/examples/command/command.cpp
+++ b/examples/command/command.cpp
@@ -163,31 +163,6 @@ std::string transcribe(whisper_context * ctx, const whisper_params & params, con
     return result;
 }
 
-// compute similarity between two strings using Levenshtein distance
-float similarity(const std::string & s0, const std::string & s1) {
-    const size_t len0 = s0.size() + 1;
-    const size_t len1 = s1.size() + 1;
-
-    std::vector<int> col(len1, 0);
-    std::vector<int> prevCol(len1, 0);
-
-    for (size_t i = 0; i < len1; i++) {
-        prevCol[i] = i;
-    }
-
-    for (size_t i = 0; i < len0; i++) {
-        col[0] = i;
-        for (size_t j = 1; j < len1; j++) {
-            col[j] = std::min(std::min(1 + col[j - 1], 1 + prevCol[j]), prevCol[j - 1] + (s0[i - 1] == s1[j - 1] ? 0 : 1));
-        }
-        col.swap(prevCol);
-    }
-
-    const float dist = prevCol[len1 - 1];
-
-    return 1.0f - (dist / std::max(s0.size(), s1.size()));
-}
-
 std::vector<std::string> read_allowed_commands(const std::string & fname) {
     std::vector<std::string> allowed_commands;
 
diff --git a/examples/common.cpp b/examples/common.cpp
index ed31efe..019a8ef 100644
--- a/examples/common.cpp
+++ b/examples/common.cpp
@@ -479,3 +479,27 @@ bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float
 
     return true;
 }
+
+float similarity(const std::string & s0, const std::string & s1) {
+    const size_t len0 = s0.size() + 1;
+    const size_t len1 = s1.size() + 1;
+
+    std::vector<int> col(len1, 0);
+    std::vector<int> prevCol(len1, 0);
+
+    for (size_t i = 0; i < len1; i++) {
+        prevCol[i] = i;
+    }
+
+    for (size_t i = 0; i < len0; i++) {
+        col[0] = i;
+        for (size_t j = 1; j < len1; j++) {
+            col[j] = std::min(std::min(1 + col[j - 1], 1 + prevCol[j]), prevCol[j - 1] + (i > 0 && s0[i - 1] == s1[j - 1] ? 0 : 1));
+        }
+        col.swap(prevCol);
+    }
+
+    const float dist = prevCol[len1 - 1];
+
+    return 1.0f - (dist / std::max(s0.size(), s1.size()));
+}
diff --git a/examples/common.h b/examples/common.h
index b08e576..48252ce 100644
--- a/examples/common.h
+++ b/examples/common.h
@@ -118,3 +118,5 @@ bool vad_simple(
         float freq_thold,
         bool  verbose);
 
+// compute similarity between two strings using Levenshtein distance
+float similarity(const std::string & s0, const std::string & s1);