feat(audio_prompts): add default audio prompts for narrator

feat(audio_prompts): add en_narrator_deep audio prompt for narrator
feat(audio_prompts): add en_narrator_light_bg audio prompt for narrator
fix(video.py): fix indentation and add prompt for generating thumbnail
fix(montage.py): fix indentation and add prompt for generating thumbnail
fix(montage.py): fix image download for wikimage slides

fix(speak.py): remove unused import statement
fix(speak.py): remove unused variable 'fakenames'
feat(speak.py): add function 'remove_blank_moments' to remove silent parts from audio file
feat(speak.py): add function 'optimize_string_groups' to optimize string groups for audio generation
fix(speak.py): fix comment indentation in 'generate_voice' function
fix(speak.py): remove unused imports in 'generate_voice' function
fix(speak.py): remove unused variable 'reduced_noise' in 'generate_voice' function
fix(speak.py): remove unused import statements in 'generate_voice' function
fix(speak.py): remove unused import statement for 'logging' module
fix(speak.py): remove unused print statements in 'main' function
fix(speak.py): remove unused import statement for 'logging' module
fix(speak.py): remove unused print statements in 'main' function
fix(speak.py):

fix(wiki_downloader.py): fix Google search URL to include correct query parameter
fix(wiki_downloader.py): reduce sleep time after page load to 1 second
fix(wiki_downloader.py): increase sleep time after image click to 5 seconds

generators/montage.py

@@ -10,7 +10,7 @@ from moviepy.editor import concatenate_videoclips, CompositeAudioClip, concatena
 from moviepy.audio.io.AudioFileClip import AudioFileClip
 from moviepy.audio.fx.all import volumex, audio_fadein, audio_fadeout # type: ignore
 from utils.misc import getenv
-
+from utils.wiki_downloader import download_image as wiki_download_image
 
 unsplash_access = getenv("unsplash_access_key")
 if not unsplash_access:
@@ -22,64 +22,82 @@ async def prepare(path):
         script = json.load(f)
         f.close()
     if not os.path.exists(path + "/slides"): os.mkdir(path + "/slides")
-    fresh = False
-    if not os.path.exists(path + "/audio"): 
-        os.mkdir(path + "/audio")
-        fresh = True
-    with open("prompts/marp.md", 'r', encoding='utf-8') as f:
+    if not os.path.exists(path + "/audio"): os.mkdir(path + "/audio") 
+    choosen_voice = random.choice(voices)
+    with open(os.path.join(os.getcwd(), "prompts", "marp.md"), 'r', encoding='utf-8') as f:
         marp = f.read()
-        f.close() 
-    if fresh:
-        choosen_voice = random.choice(voices)
-        generator = VoiceGenerator(speaker=choosen_voice)
-        for i in range(len(script)):
-            audio_path = path + "/audio/audio" + str(i) + ".wav"
-            if not os.path.exists(audio_path):
-                generator.generate_voice(audio_path, script[i]['spoken'])
-            if "image" in script[i]:
-                if os.path.exists(path + "/slides/assets/slide" + str(i) + ".md"):
-                    #skip this slide
+        f.close()
+    for i in range(len(script)):
+        audio_path = os.path.join(path, "audio", "audio" + str(i) + ".wav")
+        generator = None
+        if not os.path.exists(audio_path):
+            if not generator:
+                    generator = VoiceGenerator(speaker=choosen_voice)
+            print("Generating audio for slide " + str(i))
+            generator.generate_voice(audio_path, script[i]['spoken'])
+        if "image" in script[i]:
+            if os.path.exists(os.path.join(path, "slides", "slide" + str(i) + ".md")) and os.path.exists(os.path.join(path, "slides", "slide" + str(i) + ".png")):
+                #skip this slide
+                continue
+            if not os.path.exists(path + "/slides/assets"):
+                os.mkdir(path + "/slides/assets")
+            url= unsplash_url + script[i]['image'].replace("+", ",")
+            r = requests.get(url)
+            real_url = r.json()['urls']['raw']
+            with open(path + "/slides/assets/slide" + str(i) + ".jpg", 'wb') as f:
+                f.write(requests.get(real_url).content)
+                f.close()
+            content = marp + f"\n\n![bg 70%](assets/slide{i}.jpg)"
+            with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
+                f.write(content)
+        elif "wikimage" in script[i]:
+            if os.path.exists(os.path.join(path, "slides", "slide" + str(i) + ".md")) and os.path.exists(os.path.join(path, "slides", "slide" + str(i) + ".png")):
+                #skip this slide
+                continue
+            if not os.path.exists(path + "/slides/assets"):
+                os.mkdir(path + "/slides/assets")
+            r = 0
+            while True:
+                try: 
+                    print("Trying to download image for slide " + str(i))
+                    wiki_download_image(script[i]['wikimage'], os.path.abspath(os.path.join(path, "slides", "assets", "slide" + str(i) + ".jpg")))
+                    print("Downloaded image for slide with wikiimage " + str(i))
+                    break
+                except:
+                    r += 1
+                    if r > 5:
+                        break
                     continue
-                if not os.path.exists(path + "/slides/assets"):
-                    os.mkdir(path + "/slides/assets")
-                url= unsplash_url + script[i]['image']
-                r = requests.get(url)
-                real_url = r.json()['urls']['raw']
-                with open(path + "/slides/assets/slide" + str(i) + ".jpg", 'wb') as f:
-                    f.write(requests.get(real_url).content)
-                    f.close()
-                    content = marp + f"\n\n![bg 70%](assets/slide{i}.jpg)"
-                with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
-                    f.write(content)
-            elif "markdown" in script[i]:
-                if os.path.exists(path + "/slides/slide" + str(i) + ".md"):
-                    #skip this slide
-                    continue
-                with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
-                    f.write(marp + "\n\n" + script[i]['markdown'])
-            elif "huge" in script[i]:
-                #use fit
-                if os.path.exists(path + "/slides/slide" + str(i) + ".md"):
-                    #skip this slide
-                    continue
-                with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
-                    f.write(marp + "\n\n# <!-- fit --> " + script[i]['huge'])
-            else:
-                if os.path.exists(path + "/slides/slide" + str(i) + ".md"):
-                    #skip this slide
-                    continue
-                with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
-                    f.write(marp + "\n\n") # blank slide
+            content = marp + f"\n\n![bg 70%](assets/slide{i}.jpg)"
+            with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
+                f.write(content)
+        elif "markdown" in script[i]:
+            if os.path.exists(path + "/slides/slide" + str(i) + ".md") and os.path.exists(path + "/slides/slide" + str(i) + ".png"):
+                #skip this slide
+                continue
+            with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
+                f.write(marp + "\n\n" + script[i]['markdown'])
+        elif "huge" in script[i]:
+            #use fit
+            if os.path.exists(path + "/slides/slide" + str(i) + ".md") and os.path.exists(path + "/slides/slide" + str(i) + ".png"):
+                #skip this slide
+                continue
+            with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
+                f.write(marp + "\n\n# <!-- fit --> " + script[i]['huge'])
+        else:
+            if os.path.exists(path + "/slides/slide" + str(i) + ".md") and os.path.exists(path + "/slides/slide" + str(i) + ".png"):
+                #skip this slide
+                continue
+            with open(path + "/slides/slide" + str(i) + ".md", 'w', encoding='utf-8') as f:
+                f.write(marp + "\n\n") # blank slide
     for i in range(len(script)):
         markdown_path = os.path.join(path, f"slides/slide{i}.md")
         markdown_path = os.path.abspath(markdown_path)
         image_path = os.path.join(path, f"slides/slide{i}.png")
         image_path = os.path.abspath(image_path)
-        if os.path.exists(markdown_path):
-            #skip this slide
-            continue
-        command = f'marp.exe "{markdown_path}" -o "{image_path}" --allow-local-files'
-        os.system(command)
+        if not os.path.exists(image_path):
+            command = f'marp.exe "{markdown_path}" -o "{image_path}" --allow-local-files'
+            os.system(command)
     return script
 
 def convert_seconds_to_time_string(seconds):
@@ -113,8 +131,6 @@ async def mount(path, script):
             ])
             length = complete_audio.duration
             total_length += length
-            print(script[i])
-            print(script[i]['spoken'])
             srt = subs(length, total_length, script[i]['spoken'], srt, i)
             slide = ImageClip(path + "/slides/slide" + str(i) + ".png").set_duration(length)
             slide = slide.set_audio(complete_audio)
@@ -122,7 +138,7 @@ async def mount(path, script):
         randmusic = random.choice(os.listdir("musics"))
         while randmusic.endswith(".txt"): randmusic = random.choice(os.listdir("musics"))
         randpath = "musics/" + randmusic
-        music = AudioFileClip(randpath).set_duration(total_length)
+        music = AudioFileClip(randpath)
         music = audio_fadein(music, 20)
         music = audio_fadeout(music, 20)
         music = volumex(music, 0.2)
@@ -131,6 +147,7 @@ async def mount(path, script):
             for i in range(int(total_length / music.duration)):
                 musics.append(music)
             music = concatenate_audioclips(musics)
+        music = music.set_duration(total_length)
         final_clip = concatenate_videoclips(clips, method="compose")
         existing_audio = final_clip.audio
         final_audio = CompositeAudioClip([existing_audio, music])
@@ -142,4 +159,4 @@ async def mount(path, script):
             f.close()
         return music_credit or ""
     else:
-        return None
+        return ""

generators/speak.py

@@ -1,5 +1,5 @@
-
 import os
+from pydub import AudioSegment, silence
 
 fakenames = {
     "Alexander": "p230",
@@ -11,16 +11,70 @@ fakenames = {
 
 voices = ["Alexander", "Benjamin", "Amelia", "Katherine", "Johanne"]
 
+
+def remove_blank_moments(file_path, silence_thresh= -50, silence_chunk_len=500):
+    # Load audio file
+    audio = AudioSegment.from_wav(file_path)
+
+    # Detect non-silent parts
+    nonsilent_data = silence.detect_nonsilent(audio, min_silence_len=silence_chunk_len, silence_thresh=silence_thresh)
+
+    # Create new audio file
+    final_audio = AudioSegment.empty()
+
+    # Iterate over non-silent parts and append to the final_audio with 0.5 seconds before and after each segment
+    for idx, (start_i, end_i) in enumerate(nonsilent_data):
+        start_i = max(0, start_i - 500)  # 0.5 seconds before
+        end_i += 500  # 0.5 seconds after
+
+        segment = audio[start_i:end_i]
+
+        # Only append silence after the first segment
+        if idx > 0:
+            final_audio += AudioSegment.silent(duration=500)
+
+        final_audio += segment
+    # Save the result
+    if not os.path.exists(os.path.abspath(os.path.join(os.getcwd(), "temp"))):
+        os.mkdir(os.path.abspath(os.path.join(os.getcwd(), "temp")))
+    tempfile_path = os.path.abspath(os.path.join(os.getcwd(), "temp", "temp.wav"))
+    final_audio.export(tempfile_path, format="wav")
+    os.remove(file_path)
+    os.rename(tempfile_path, file_path)
+
+
+def optimize_string_groups(strings):
+    optimized_groups = []
+    current_group = []
+    current_length = 0
+
+    for string in strings:
+        string_length = len(string) + len(current_group)  # Account for spaces between strings
+        if current_length + string_length <= 100:
+            current_group.append(string)
+            current_length += string_length
+        else:
+            optimized_groups.append(' '.join(current_group))  # Join strings with spaces
+            current_group = [string]
+            current_length = len(string)
+
+    if current_group:
+        optimized_groups.append(' '.join(current_group))
+
+    return optimized_groups
+
 class VoiceGenerator:
     def __init__(self, mode="Bark", speaker=""):
         self.mode = mode
         self.speaker = speaker
         if mode == "Bark":
             os.environ["XDG_CACHE_HOME"] = os.path.join(os.getcwd(), "bark_cache")
-            from bark import preload_models, generation
-
+            from bark import preload_models
+            print("Loading Bark voice generator")
             preload_models()
-            self.speaker = "v2/en_speaker_6"
+            #self.speaker = os.path.abspath(os.path.join(os.getcwd(), "audio_prompts", "en_male_professional_reader.npz"))
+            self.speaker = os.path.join(os.getcwd(), "audio_prompts", "en_narrator_light_bg.npz")
+            print(f"Generating voice for Bark with speaker {self.speaker}")
         else:
             from TTS.api import TTS
             model = "tts_models/en/vctk/vits"
@@ -43,20 +97,27 @@ class VoiceGenerator:
             import numpy as np
             import nltk
             sentences = nltk.sent_tokenize(text)
+            sentences = optimize_string_groups(sentences)
+            print(sentences)
             pieces = []
             silence = np.zeros(int(0.25 * SAMPLE_RATE))  # quarter second of silence
             for sentence in sentences:
-                audio_array = generate_audio(sentence, history_prompt=self.speaker)
-                pieces += [audio_array, silence.copy()]
+                if not sentence == "":
+                    audio_array = generate_audio(sentence, history_prompt=self.speaker)
+                    pieces += [audio_array, silence.copy()]
             audio_array = np.concatenate(pieces)
             soundfile.write(path, audio_array, SAMPLE_RATE, format="WAV", subtype="PCM_16")
-            rate, data = wavread(path)
-            reduced_noise = nr.reduce_noise(y=data, sr=rate)
-            os.remove(path)
-            wavwrite(path, rate, reduced_noise)
+            '''
+            remove  silence
+            '''
+            remove_blank_moments(path)
         else:
             self.tts.tts_to_file(text=text, file_path=path, speaker=self.speaker, speed=1, emotion="Happy")
 if __name__ == "__main__":
+    import logging
+    logging.basicConfig(level=logging.INFO)
+    print("Testing voice generator")
     generator = VoiceGenerator()
-    generator.generate_voice("test/test_r.wav", "Hello there!")
-    generator.generate_voice("test/teste_r.wav", "This is a test. I like the words python, django and flask. Betty bought a bit of butter but the butter was bitter. So she bought some better butter to make the bitter butter better.")
+    print("Loaded voice generator")
+#    generator.generate_voice("test/test_r.wav", "Hello there!")
+    generator.generate_voice("test/tast_timbernerslee.wav", "But his greatest claim to fame is undoubtedly his invention of the World Wide Web back in 1989. Can you imagine a world without the internet? [Laughs] No, thank you!")

generators/thumbnail.py

@@ -28,6 +28,11 @@ Answer without anything else, just with the 2 textes. Answer with text1 on the f
 Here is the title of the video: [TITLE]
 Here is the description of the video: [DESCRIPTION]'''
 
+
+# TODO: make jpg qith 90% quality default when generating the image to avoid having to convert it later
+
+
+
 async def rand_gradient(image):
     randr = random.SystemRandom().randint(1, 20)
     randg = random.SystemRandom().randint(1, 20)
@@ -110,11 +115,20 @@ async def generate_image(path, text1, text2):
     drawtext2.text((imgtext2.size[0]//8*2.5, imgtext2.size[1]//5*2), text2def, font=font2, fill=(textcolor2[0], textcolor2[1], textcolor2[2]))
     imgtext2 = imgtext2.rotate(5, expand=True)
     #paste the textes on the image
-    img.paste(bcg, (0, 0), bcg)
+    bcg = bcg.convert('RGBA')
+    #also set the bcg size to the image size
+    bcg = bcg.resize((1920, 1080))
+    img.paste(bcg, (0, 0), bcg) # TODO: make it work with standard pngs (non rgba)
     img.paste(imgtext1, (0, 0-img.size[1]//8), imgtext1)
     if len(text1def.split("\n")) > 2: #if the text is too long, put the second text on the third line
         img.paste(imgtext2, (0, img.size[1]//8), imgtext2)
     else:
         img.paste(imgtext2, (0, 0), imgtext2)
-    img.save(path + "/miniature.png")
-    return path + "/miniature.png"
+    #disable the alpha channel
+    img = img.convert('RGB')
+    img_path = os.path.abspath(os.path.join(path, "thumbnail.jpg"))
+    for quality in range(100, 0, -1):
+        img.save(img_path, quality=quality)
+        if os.path.getsize(img_path) < 2000000:
+            break
+    return img_path