1+ import io
2+ import wave
3+ import hashlib
4+
5+ from vocode import getenv
6+ from vocode .streaming .models .audio import AudioEncoding , SamplingRate
7+ from vocode .streaming .models .message import BaseMessage
8+ from vocode .streaming .models .synthesizer import CartesiaSynthesizerConfig
9+ from vocode .streaming .synthesizer .base_synthesizer import BaseSynthesizer , SynthesisResult
10+ from vocode .streaming .utils .create_task import asyncio_create_task_with_done_error_log
11+
12+
13+ class CartesiaSynthesizer (BaseSynthesizer [CartesiaSynthesizerConfig ]):
14+ def __init__ (
15+ self ,
16+ synthesizer_config : CartesiaSynthesizerConfig ,
17+ ):
18+ super ().__init__ (synthesizer_config )
19+
20+ # Lazy import the cartesia module
21+ try :
22+ from cartesia .tts import AsyncCartesiaTTS
23+ except ImportError as e :
24+ raise ImportError (
25+ f"Missing required dependancies for CartesiaSynthesizer"
26+ ) from e
27+
28+ self .cartesia_tts = AsyncCartesiaTTS
29+
30+ self .api_key = synthesizer_config .api_key or getenv ("CARTESIA_API_KEY" )
31+ if not self .api_key :
32+ raise ValueError ("Missing Cartesia API key" )
33+
34+
35+ if synthesizer_config .audio_encoding == AudioEncoding .LINEAR16 :
36+ self .channel_width = 2
37+ match synthesizer_config .sampling_rate :
38+ case SamplingRate .RATE_44100 :
39+ self .sampling_rate = 44100
40+ self .output_format = "pcm_44100"
41+ case SamplingRate .RATE_22050 :
42+ self .sampling_rate = 22050
43+ self .output_format = "pcm_22050"
44+ case SamplingRate .RATE_16000 :
45+ self .sampling_rate = 16000
46+ self .output_format = "pcm_16000"
47+ case _:
48+ raise ValueError (
49+ f"Unsupported PCM sampling rate { synthesizer_config .sampling_rate }
50+ )
51+ elif synthesizer_config .audio_encoding == AudioEncoding .MULAW :
52+ # Cartesia has issues with MuLaw/8000. Use pcm/16000 and
53+ # create_synthesis_result_from_wav will handle the conversion to mulaw/8000
54+ self .channel_width = 2
55+ self .output_format = "pcm_16000"
56+ self .sampling_rate = 16000
57+ else :
58+ raise ValueError (
59+ f"Unsupported audio encoding { synthesizer_config .audio_encoding }
60+ )
61+
62+ self .num_channels = 1
63+ self .model_id = synthesizer_config .model_id
64+ self .voice_id = synthesizer_config .voice_id
65+ self .client = self .cartesia_tts (api_key = self .api_key )
66+ self .voice_embedding = self .client .get_voice_embedding (voice_id = self .voice_id )
67+
68+
69+ async def create_speech_uncached (
70+ self ,
71+ message : BaseMessage ,
72+ chunk_size : int ,
73+ is_first_text_chunk : bool = False ,
74+ is_sole_text_chunk : bool = False ,
75+ ) -> SynthesisResult :
76+ generator = await self .client .generate (
77+ transcript = message .text ,
78+ voice = self .voice_embedding ,
79+ stream = True ,
80+ model_id = self .model_id ,
81+ data_rtype = 'bytes' ,
82+ output_format = self .output_format
83+ )
84+
85+ audio_file = io .BytesIO ()
86+ with wave .open (audio_file , 'wb' ) as wav_file :
87+ wav_file .setnchannels (self .num_channels )
88+ wav_file .setsampwidth (self .channel_width )
89+ wav_file .setframerate (self .sampling_rate )
90+ async for chunk in generator :
91+ wav_file .writeframes (chunk ['audio' ])
92+ audio_file .seek (0 )
93+
94+ result = self .create_synthesis_result_from_wav (
95+ synthesizer_config = self .synthesizer_config ,
96+ file = audio_file ,
97+ message = message ,
98+ chunk_size = chunk_size ,
99+ )
100+
101+ return result
102+
103+ @classmethod
104+ def get_voice_identifier (cls , synthesizer_config : CartesiaSynthesizerConfig ):
105+ hashed_api_key = hashlib .sha256 (f"{ synthesizer_config .api_key } .encode ("utf-8" )).hexdigest ()
106+ return ":" .join (
107+ (
108+ "cartesia" ,
109+ hashed_api_key ,
110+ str (synthesizer_config .voice_id ),
111+ str (synthesizer_config .model_id ),
112+ synthesizer_config .audio_encoding
113+ )
114+ )
0 commit comments