add melo synthesizer

bolna/constants.py

@@ -32,4 +32,12 @@
     "that's enough", "shush", "listen", "excuse me", "hold up", "not now", "stop there", "stop speaking"
 ]
 
-PRE_FUNCTIONAL_CALL_MESSAGE = "Just give me a moment, I'll do that and will be back with you"
+PRE_FUNCTIONAL_CALL_MESSAGE = "Just give me a moment, I'll do that and will be back with you"
+
+FILLER_PHRASES = [
+    "No worries.", "It's fine.", "I'm here.", "No rush.", "Take your time.",
+    "Great!", "Awesome!", "Fantastic!", "Wonderful!", "Perfect!", "Excellent!",
+    "I get it.", "Noted.", "Alright.", "I understand.", "Understood.", "Got it.",
+    "Sure.", "Okay.", "Right.", "Absolutely.", "Sure thing.",
+    "I see.", "Gotcha.", "Makes sense."
+]

bolna/models.py

@@ -47,6 +47,13 @@ class FourieConfig(BaseModel):
 class DeepgramConfig(BaseModel):
     voice: str
 
+class MeloConfig(BaseModel):
+    voice:str
+    sample_rate:int
+    sdp_ratio:float = 0.2
+    noise_scale:float = 0.6
+    noise_scale_w:float =  0.8
+    speed:float = 1.0
 
 class Transcriber(BaseModel):
     model: str
@@ -70,15 +77,15 @@ def validate_language(cls, value):
 
 class Synthesizer(BaseModel):
     provider: str
-    provider_config: Union[PollyConfig, XTTSConfig, ElevenLabsConfig, OpenAIConfig, FourieConfig, DeepgramConfig]
+    provider_config: Union[PollyConfig, XTTSConfig, ElevenLabsConfig, OpenAIConfig, FourieConfig, MeloConfig, DeepgramConfig]
     stream: bool = False
     buffer_size: Optional[int] = 40  # 40 characters in a buffer
     audio_format: Optional[str] = "pcm"
     caching: Optional[bool] = True
 
     @validator("provider")
     def validate_model(cls, value):
-        return validate_attribute(value, ["polly", "xtts", "elevenlabs", "openai", "deepgram"])
+        return validate_attribute(value, ["polly", "xtts", "elevenlabs", "openai", "deepgram", "meloTTS"])
 
 
 class IOModel(BaseModel):
@@ -181,6 +188,9 @@ class Task(BaseModel):
     task_type: Optional[str] = "conversation"  # extraction, summarization, notification
     task_config: ConversationConfig = dict()
 
+    @validator('voice')
+    def check_voice(cls, value):
+        return validate_attribute(value, ['Jason','Oscar','Travis','Joseph','Ram'])
 
 class AgentModel(BaseModel):
     agent_name: str

bolna/providers.py

@@ -1,4 +1,4 @@
-from .synthesizer import PollySynthesizer, XTTSSynthesizer, ElevenlabsSynthesizer, OPENAISynthesizer, FourieSynthesizer, DeepgramSynthesizer
+from .synthesizer import PollySynthesizer, XTTSSynthesizer, ElevenlabsSynthesizer, OPENAISynthesizer, FourieSynthesizer, DeepgramSynthesizer, MeloSynthesizer
 from .transcriber import DeepgramTranscriber, WhisperTranscriber
 from .input_handlers import DefaultInputHandler, TwilioInputHandler, ExotelInputHandler, PlivoInputHandler
 from .output_handlers import DefaultOutputHandler, TwilioOutputHandler, ExotelOutputHandler, PlivoOutputHandler
@@ -10,7 +10,8 @@
     'elevenlabs': ElevenlabsSynthesizer,
     'openai': OPENAISynthesizer,
     'fourie': FourieSynthesizer,
-    'deepgram': DeepgramSynthesizer
+    'deepgram': DeepgramSynthesizer,
+    'meloTTS': MeloSynthesizer
 }
 SUPPORTED_TRANSCRIBER_MODELS = {
     'deepgram': DeepgramTranscriber,

bolna/synthesizer/__init__.py

@@ -4,3 +4,4 @@
 from .openai_synthesizer import OPENAISynthesizer
 from .fourie_synthesizer import FourieSynthesizer
 from .deepgram_synthesizer import DeepgramSynthesizer
+from .melo_synthesizer import MeloSynthesizer

bolna/synthesizer/melo_synthesizer.py

@@ -0,0 +1,83 @@
+
+import aiohttp
+import os
+from dotenv import load_dotenv
+from bolna.helpers.logger_config import configure_logger
+from bolna.helpers.utils import create_ws_data_packet
+from .base_synthesizer import BaseSynthesizer
+import json
+import base64
+
+load_dotenv()
+logger = configure_logger(__name__)
+
+
+
+class MeloSynthesizer(BaseSynthesizer):
+    def __init__(self, audio_format="pcm", sampling_rate="8000", stream=False, buffer_size=400,
+                 **kwargs):
+        super().__init__(stream, buffer_size)
+        self.format = "linear16" if audio_format == "pcm" else audio_format
+        self.sample_rate = int(sampling_rate)
+        self.first_chunk_generated = False
+        self.url = os.getenv('MELO_TTS')
+
+        self.voice = kwargs.get('voice')
+        self.sample_rate = kwargs.get('sample_rate')
+        self.sdp_ratio = kwargs.get('sdp_ratio')
+        self.noise_scale=kwargs.get('noise_scale')
+        self.noise_scale_w = kwargs.get('noise_scale_w')
+        self.speed = kwargs.get('speed')
+        # self.voice_id
+
+    async def __generate_http(self, text):
+        payload = {
+            "voice_id": self.voice,
+            "text": text,
+            "sr": self.sample_rate,
+            "sdp_ratio" : self.sdp_ratio,
+            "noise_scale" : self.noise_scale,
+            "noise_scale_w" :  self.noise_scale_w,
+            "speed" : self.speed
+        }
+
+        headers = {
+            'Content-Type': 'application/json'
+        }
+
+        async with aiohttp.ClientSession() as session:
+            if payload is not None:
+                async with session.post(self.url, headers=headers, json=payload) as response:
+                    if response.status == 200:
+                        res_json:dict = json.loads(await response.text())
+                        chunk = base64.b64decode(res_json["audio"])
+                        yield chunk
+            else:
+                logger.info("Payload was null")
+
+    async def open_connection(self):
+        pass
+
+    async def generate(self):
+        while True:
+            message = await self.internal_queue.get()
+            logger.info(f"Generating TTS response for message: {message}")
+
+            meta_info, text = message.get("meta_info"), message.get("data")
+            async for message in self.__generate_http(text):
+                if not self.first_chunk_generated:
+                    meta_info["is_first_chunk"] = True
+                    self.first_chunk_generated = True
+                else:
+                    meta_info["is_first_chunk"] = False
+                if "end_of_llm_stream" in meta_info and meta_info["end_of_llm_stream"]:
+                    meta_info["end_of_synthesizer_stream"] = True
+                    self.first_chunk_generated = False
+
+                meta_info['text'] = text
+                meta_info['format'] = self.format
+                yield create_ws_data_packet(message, meta_info)
+
+    async def push(self, message):
+        logger.info("Pushed message to internal queue")
+        self.internal_queue.put_nowait(message)