Update script to do smarter sentence breaking and joining.

speak.py

@@ -3,25 +3,46 @@ from urllib.parse import quote
 import subprocess
 import os
 import sys
+import re
 
 speaker_id="p230"
 
-letter_year = sys.argv[1]
+letter_series = sys.argv[1] or "buffett-partnership"
+letter_year = sys.argv[2] or "1956"
+real_run = (len(sys.argv) == 4) and bool(sys.argv[3]) or False
 
-fn = f"./annual-letters/bpl-{letter_year}.txt"
+fn = f"./{letter_series}/{letter_year}/1-in/bpl-{letter_year}-letter.txt"
+pre_fn = f"./{letter_series}/{letter_year}/2-preprocess/bpl-{letter_year}.txt"
 
-output_dir = f"./wavs/bpl-{letter_year}/"
+output_dir = f"./{letter_series}/{letter_year}/3-wavs"
 
 subprocess.run(['mkdir', '-p', output_dir])
 
+if (not real_run):
+    print("Dry run, check that sentences are readable.")
+
+def synth_sentence(sentence, i):
+    text_prompt = quote(sentence)
+    # If hosted on IPv6 use http://[::1]:5002
+    # If hosted on IPv4 use http://localhost:5002
+    query_string = f"http://\[::1\]:5002/api/tts?text={text_prompt}&style_wav=&language_id=&speaker_id={speaker_id}"
+    print(f"Query string {query_string}")
+    subprocess.run(['curl', query_string, '-o', f"./{output_dir}/output-{str(i).zfill(3)}.wav"])
+
 with open(fn) as f:
     lines = f.readlines()
     all_text = ''.join(lines)
-    sentences = all_text.split('.')
-    count = len(sentences)
-
-    for (i, sentence) in enumerate(sentences):
-        text_prompt = quote(sentence)
-        query_string = f"http://localhost:5002/api/tts?text={text_prompt}&speaker_id={speaker_id}&style_wav=&language_id="
-        subprocess.run(['curl', query_string, '-o', f"./{output_dir}/output-{str(i).zfill(3)}.wav"])
-
+    sentence_ends = list(re.finditer(r"(\w\w+([\.:][\s\n])|([\s\n][\s\n]+))", all_text))
+    
+    start = 0
+    count = len(sentence_ends)
+    print(f"{count} sentences found.\n")
+
+    for (i, sentence_end) in enumerate(sentence_ends):
+        sentence = all_text[start:sentence_end.end()].strip()
+        # Heal any newlines due to awkward breaks from copy-pasta
+        sentence = " ".join(map(lambda x: x.strip(), sentence.split("\n")))
+        print(f"Sentence {i}:\n\t\"{sentence}\"\n\t{sentence_end}")
+        start = sentence_end.end() # no +1 needed since end is already one past the real ending character
+        if (real_run):
+            synth_sentence(sentence, i)