HFA: fix fill small gap

Author: wolfgitpr

src/apps/HubertFA/util/AlignWord.cpp

@@ -23,21 +23,22 @@ namespace HFA {
         return a >= b - EPS;
     }
 
-    Phoneme::Phoneme(const float start, const float end, const std::string &text)
-        : start(std::max(0.0f, start)), end(std::max(0.0f, end)), text(text) {
+    Phoneme::Phoneme(const float p_start, const float p_end, const std::string &text)
+        : start(std::max(0.0f, p_start)), end(std::max(0.0f, p_end)), text(text) {
         if (!(0 <= start && start < end)) {
-            std::cerr << "Warning: Phoneme Invalid: text=" << text << " start=" << start << ", end=" << end
+            std::cerr << "Warning: Phoneme Invalid: text=" << text << " p_start=" << start << ", p_end=" << end
                       << std::endl;
         }
     }
 
-    Word::Word(float start, float end, const std::string &text, const bool init_phoneme)
-        : start(std::max(0.0f, start)), end(std::max(0.0f, end)), text(text) {
+    Word::Word(float w_start, float w_end, const std::string &text, const bool init_phoneme)
+        : start(std::max(0.0f, w_start)), end(std::max(0.0f, w_end)), text(text) {
         if (!(0 <= start && start < end)) {
-            std::cerr << "Warning: Word Invalid: text=" << text << " start=" << start << ", end=" << end << std::endl;
+            std::cerr << "Warning: Word Invalid: text=" << text << " w_start=" << start << ", w_end=" << end
+                      << std::endl;
         }
         if (init_phoneme) {
-            phonemes.emplace_back(start, end, text);
+            phonemes.emplace_back(w_start, w_end, text);
         }
     }
 

src/apps/HubertFA/util/AlignWord.h

@@ -12,7 +12,7 @@ namespace HFA {
         float end;
         std::string text;
 
-        Phoneme(float start, float end, const std::string &text);
+        Phoneme(float p_start, float p_end, const std::string &text);
     };
 
     class Word {
@@ -22,7 +22,7 @@ namespace HFA {
         std::string text;
         std::vector<Phoneme> phonemes;
 
-        Word(float start, float end, const std::string &text, bool init_phoneme = false);
+        Word(float w_start, float w_end, const std::string &text, bool init_phoneme = false);
 
         float dur() const;
         void add_phoneme(const Phoneme &phoneme);

src/apps/HubertFA/util/AlignmentDecoder.cpp

@@ -208,9 +208,6 @@ namespace HFA {
                 word_idx_last = word_idx;
             }
         }
-
-        // 填充小间隙
-        words.fill_small_gaps(wav_length);
         return true;
     }
 

src/apps/HubertFA/util/Hfa.cpp

@@ -79,7 +79,7 @@ namespace HFA {
         sf.seek(0, SEEK_SET);
         sf.read(audio.data(), static_cast<sf_count_t>(audio.size()));
 
-        const auto wav_length = static_cast<double>(sf.frames()) / hfa_input_sample_rate;
+        const float wav_length = static_cast<float>(sf.frames()) / hfa_input_sample_rate;
 
         if (wav_length > 60) {
             msg = "The audio contains continuous pronunciation segments that exceed 60 seconds. Please manually "
@@ -129,6 +129,7 @@ namespace HFA {
                 for (const auto &word : word_list)
                     words.add_AP(word);
 
+            words.fill_small_gaps(wav_length);
             words.clear_language_prefix();
             words.add_SP(wav_length, "SP");
             return true;

src/apps/HubertFA/util/HfaThread.cpp

@@ -39,9 +39,10 @@ namespace HFA {
                 auto tierPhones = std::make_shared<textgrid::IntervalTier>("phonemes", 0.0, words.duration());
 
                 for (auto word : words) {
-                    tierWords->AppendInterval(textgrid::Interval(word->start, word->end, word->text));
+                    tierWords->AppendInterval(textgrid::Interval(std::max(0.0f, word->start), word->end, word->text));
                     for (auto phoneme : word->phonemes) {
-                        tierPhones->AppendInterval(textgrid::Interval(phoneme.start, phoneme.end, phoneme.text));
+                        tierPhones->AppendInterval(
+                            textgrid::Interval(std::max(0.0f, phoneme.start), phoneme.end, phoneme.text));
                     }
                 }