Text Chunking for TTS REST Optimization

1
from deepgram import (
2
    DeepgramClient,
3
    SpeakOptions,
4
)
5
6
# Define the maximum chunk size (in characters) for text chunking
7
MAX_CHUNK_SIZE = 200
8
9
input_text = "Our story begins in a peaceful woodland kingdom where a lively squirrel named Frolic made his abode high up within a cedar tree's embrace. He was not a usual woodland creature, for he was blessed with an insatiable curiosity and a heart for adventure. Nearby, a glistening river snaked through the landscape, home to a wonder named Splash - a silver-scaled flying fish whose ability to break free from his water-haven intrigued the woodland onlookers. This magical world moved on a rhythm of its own until an unforeseen circumstance brought Frolic and Splash together. One radiant morning, while Frolic was on his regular excursion, and Splash was making his aerial tours, an unpredictable wave playfully tossed and misplaced Splash onto the riverbank. Despite his initial astonishment, Frolic hurriedly and kindly assisted his new friend back to his watery abode. Touched by Frolic's compassion, Splash expressed his gratitude by inviting his friend to share his world. As Splash perched on Frolic's back, he tasted of the forest's bounty, felt the sun’s rays filter through the colors of the trees, experienced the conversations amidst the woods, and while at it, taught the woodland how to blur the lines between earth and water."
10
11
def chunk_text(text, chunk_size):
12
    chunks = []
13
    words = text.split()
14
    current_chunk = ''
15
    for word in words:
16
        if len(current_chunk) + len(word) <= chunk_size:
17
            current_chunk += ' ' + word
18
        else:
19
            chunks.append(current_chunk.strip())
20
            current_chunk = word
21
    if current_chunk:
22
        chunks.append(current_chunk.strip())
23
    return chunks
24
25
def main():
26
    try:
27
        # Create a Deepgram client using the API key
28
        deepgram = DeepgramClient(api_key="DEEPGRAM_API_KEY")
29
30
        # Choose a model to use for synthesis
31
        options = SpeakOptions(
32
            model="aura-helios-en",
33
        )
34
35
        # Chunk the text into smaller parts
36
        text_chunks = chunk_text(input_text, MAX_CHUNK_SIZE)
37
        
38
        # Synthesize audio for each chunk
39
        for i, chunk in enumerate(text_chunks):
40
            print(f"\nProcessing chunk {i + 1}...{chunk}\n")
41
            filename = f"chunk_{i + 1}.mp3"
42
43
            SPEAK_OPTIONS = {"text": chunk}
44
        
45
            response = deepgram.speak.v("1").save(filename, SPEAK_OPTIONS, options)
46
            print(response.to_json(indent=4))
47
48
    except Exception as e:
49
        print(f"Exception: {e}")
50
51
if __name__ == "__main__":
52
    main()

1
import re
2
from deepgram import (
3
    DeepgramClient,
4
    SpeakOptions,
5
)
6
7
input_text = "Our story begins in a peaceful woodland kingdom where a lively squirrel named Frolic made his abode high up within a cedar tree's embrace. He was not a usual woodland creature, for he was blessed with an insatiable curiosity and a heart for adventure. Nearby, a glistening river snaked through the landscape, home to a wonder named Splash - a silver-scaled flying fish whose ability to break free from his water-haven intrigued the woodland onlookers. This magical world moved on a rhythm of its own until an unforeseen circumstance brought Frolic and Splash together. One radiant morning, while Frolic was on his regular excursion, and Splash was making his aerial tours, an unpredictable wave playfully tossed and misplaced Splash onto the riverbank. Despite his initial astonishment, Frolic hurriedly and kindly assisted his new friend back to his watery abode. Touched by Frolic's compassion, Splash expressed his gratitude by inviting his friend to share his world. As Splash perched on Frolic's back, he tasted of the forest's bounty, felt the sun’s rays filter through the colors of the trees, experienced the conversations amidst the woods, and while at it, taught the woodland how to blur the lines between earth and water."
8
9
CLAUSE_BOUNDARIES = r'\.|\?|!|;|, (and|but|or|nor|for|yet|so)'
10
11
def chunk_text_by_clause(text):
12
    # Find clause boundaries using regular expression
13
    clause_boundaries = re.finditer(CLAUSE_BOUNDARIES, text)
14
    boundaries_indices = [boundary.start() for boundary in clause_boundaries]
15
16
    chunks = []
17
    start = 0
18
    for boundary_index in boundaries_indices:
19
        chunks.append(text[start:boundary_index + 1].strip())
20
        start = boundary_index + 1
21
    # Append the remaining part of the text
22
    chunks.append(text[start:].strip())
23
24
    return chunks
25
26
def main():
27
    try:
28
        # Create a Deepgram client using the API key
29
        deepgram = DeepgramClient(api_key="DEEPGRAM_API_KEY")
30
31
        # Choose a model to use for synthesis
32
        options = SpeakOptions(
33
            model="aura-helios-en",
34
        )
35
36
        # Chunk the text into smaller parts
37
        text_chunks = chunk_text_by_clause(input_text)
38
        
39
        # Synthesize audio for each chunk
40
        for i, chunk in enumerate(text_chunks):
41
            print(f"\nProcessing chunk {i + 1}...{chunk}\n")
42
            filename = f"chunk_{i + 1}.mp3"
43
44
            SPEAK_OPTIONS = {"text": chunk}
45
        
46
            response = deepgram.speak.v("1").save(filename, SPEAK_OPTIONS, options)
47
            print(response.to_json(indent=4))
48
49
    except Exception as e:
50
        print(f"Exception: {e}")
51
52
if __name__ == "__main__":
53
    main()

1
import re
2
from deepgram import (
3
    DeepgramClient,
4
    SpeakOptions,
5
)
6
7
input_text = "Our story begins in a peaceful woodland kingdom where a lively squirrel named Frolic made his abode high up within a cedar tree's embrace. He was not a usual woodland creature, for he was blessed with an insatiable curiosity and a heart for adventure. Nearby, a glistening river snaked through the landscape, home to a wonder named Splash - a silver-scaled flying fish whose ability to break free from his water-haven intrigued the woodland onlookers. This magical world moved on a rhythm of its own until an unforeseen circumstance brought Frolic and Splash together. One radiant morning, while Frolic was on his regular excursion, and Splash was making his aerial tours, an unpredictable wave playfully tossed and misplaced Splash onto the riverbank. Despite his initial astonishment, Frolic hurriedly and kindly assisted his new friend back to his watery abode. Touched by Frolic's compassion, Splash expressed his gratitude by inviting his friend to share his world. As Splash perched on Frolic's back, he tasted of the forest's bounty, felt the sun’s rays filter through the colors of the trees, experienced the conversations amidst the woods, and while at it, taught the woodland how to blur the lines between earth and water."
8
9
CLAUSE_BOUNDARIES = r'\.|\?|!|;|, (and|but|or|nor|for|yet|so)'
10
MAX_CHUNK_LENGTH = 100
11
12
def chunk_text_dynamically(text):
13
    # Find clause boundaries using regular expression
14
    clause_boundaries = re.finditer(CLAUSE_BOUNDARIES, text)
15
    boundaries_indices = [boundary.start() for boundary in clause_boundaries]
16
17
    chunks = []
18
    start = 0
19
    # Add chunks until the last clause boundary
20
    for boundary_index in boundaries_indices:
21
        chunk = text[start:boundary_index + 1].strip()
22
        if len(chunk) <= MAX_CHUNK_LENGTH:
23
            chunks.append(chunk)
24
        else:
25
            # Split by comma if it doesn't create subchunks less than three words
26
            subchunks = chunk.split(',')
27
            temp_chunk = ''
28
            for subchunk in subchunks:
29
                if len(temp_chunk) + len(subchunk) <= MAX_CHUNK_LENGTH:
30
                    temp_chunk += subchunk + ','
31
                else:
32
                    if len(temp_chunk.split()) >= 3:
33
                        chunks.append(temp_chunk.strip())
34
                    temp_chunk = subchunk + ','
35
            if temp_chunk:
36
                if len(temp_chunk.split()) >= 3:
37
                    chunks.append(temp_chunk.strip())
38
        start = boundary_index + 1
39
40
    # Split remaining text into subchunks if needed
41
    remaining_text = text[start:].strip()
42
    if remaining_text:
43
        remaining_subchunks = [remaining_text[i:i+MAX_CHUNK_LENGTH] for i in range(0, len(remaining_text), MAX_CHUNK_LENGTH)]
44
        chunks.extend(remaining_subchunks)
45
46
    return chunks
47
48
def main():
49
    try:
50
        # Create a Deepgram client using the API key
51
        deepgram = DeepgramClient(api_key="DEEPGRAM_API_KEY")
52
53
        # Choose a model to use for synthesis
54
        options = SpeakOptions(
55
            model="aura-helios-en",
56
        )
57
58
        # Chunk the text into smaller parts
59
        text_chunks = chunk_text_dynamically(input_text)
60
        
61
        # Synthesize audio for each chunk
62
        for i, chunk in enumerate(text_chunks):
63
            print(f"\nProcessing chunk {i + 1}...{chunk}\n")
64
            filename = f"chunk_{i + 1}.mp3"
65
66
            SPEAK_OPTIONS = {"text": chunk}
67
        
68
            response = deepgram.speak.v("1").save(filename, SPEAK_OPTIONS, options)
69
            # print(response.to_json(indent=4))
70
71
    except Exception as e:
72
        print(f"Exception: {e}")
73
74
if __name__ == "__main__":
75
    main()

1
import re
2
from deepgram import (
3
    DeepgramClient,
4
    SpeakOptions,
5
)
6
from pydub import AudioSegment
7
from pydub.playback import play
8
9
input_text = "Our story begins in a peaceful woodland kingdom where a lively squirrel named Frolic made his abode high up within a cedar tree's embrace. He was not a usual woodland creature, for he was blessed with an insatiable curiosity and a heart for adventure. Nearby, a glistening river snaked through the landscape, home to a wonder named Splash - a silver-scaled flying fish whose ability to break free from his water-haven intrigued the woodland onlookers. This magical world moved on a rhythm of its own until an unforeseen circumstance brought Frolic and Splash together. One radiant morning, while Frolic was on his regular excursion, and Splash was making his aerial tours, an unpredictable wave playfully tossed and misplaced Splash onto the riverbank. Despite his initial astonishment, Frolic hurriedly and kindly assisted his new friend back to his watery abode. Touched by Frolic's compassion, Splash expressed his gratitude by inviting his friend to share his world. As Splash perched on Frolic's back, he tasted of the forest's bounty, felt the sun’s rays filter through the colors of the trees, experienced the conversations amidst the woods, and while at it, taught the woodland how to blur the lines between earth and water."
10
11
def chunk_text_by_sentence(text):
12
    # Find sentence boundaries using regular expression
13
    sentence_boundaries = re.finditer(r'(?<=[.!?])\s+', text)
14
    boundaries_indices = [boundary.start() for boundary in sentence_boundaries]
15
16
    chunks = []
17
    start = 0
18
    # Split the text into chunks based on sentence boundaries
19
    for boundary_index in boundaries_indices:
20
        chunks.append(text[start:boundary_index + 1].strip())
21
        start = boundary_index + 1
22
    chunks.append(text[start:].strip())
23
24
    return chunks
25
26
def synthesize_audio(text):
27
    # Create a Deepgram client using the API key
28
    deepgram = DeepgramClient(api_key="DEEPGRAM_API_KEY")
29
    # Choose a model to use for synthesis
30
    options = SpeakOptions(
31
            model="aura-helios-en",
32
        )
33
    speak_options = {"text": text}
34
    # Synthesize audio and stream the response
35
    response =  deepgram.speak.v("1").stream(speak_options, options)
36
    # Get the audio stream from the response
37
    audio_buffer = response.stream
38
    audio_buffer.seek(0)
39
    # Load audio from buffer using pydub
40
    audio = AudioSegment.from_mp3(audio_buffer)
41
42
    return audio
43
44
def main():
45
    # Chunk the text into smaller parts
46
    chunks = chunk_text_by_sentence(input_text)
47
48
    # Synthesize each chunk into audio and play the audio
49
    for chunk_text in chunks:
50
        audio = synthesize_audio(chunk_text)
51
        play(audio)
52
53
if __name__ == "__main__":
54
    main()