Class: Insika::Tools::Tts

Inherits:
RubyLLM::Tool
  • Object
show all
Defined in:
lib/insika/tools/tts.rb

Overview

The agent's SPEECH output (WS9, saída). The engine transports media, never meaning: the words belong to the customer's channel (a voice note on WhatsApp), and the contract here just produces the audio bytes and carries them in the turn's output_parts.

Wired ONLY when both gates pass (ChatBuilder): the agent opted in (outputs.tts) AND the channel declared it can receive the media (channel.capabilities includes "audio_output") — nothing leaks by default. The clip is an envelope part, never part of the answer text; the turn counts the call in its usage (usage.media) — the provider's speech API reports no token counts, so the part carries the model for consumer-side pricing.

Instance Method Summary collapse

Constructor Details

#initialize(runner:, config:, state:) ⇒ Tts

runner: a duck exposing #generate_media_output(:tts, text, config) -> [part, usage] and #account_media_usage(part, usage) (the Executor).



31
32
33
34
35
36
 
# File 'lib/insika/tools/tts.rb', line 31

def initialize(runner:, config:, state:, **)
  @runner = runner
  @config = config
  @state = state
  super()
end

Instance Method Details

#execute(text:, voice: nil) ⇒ Object 



38
39
40
41
42
43
44
 
# File 'lib/insika/tools/tts.rb', line 38

def execute(text:, voice: nil)
  cfg = @config.merge("voice" => voice.to_s).reject { |_, v| v.to_s.empty? }
  part, usage = @runner.generate_media_output(:tts, text.to_s, cfg)
  @state.output_parts << part
  @runner.(@state, part, usage)
  "speech synthesized and attached to the reply (#{part["mime_type"]})"
end

#name ⇒ Object 



27
 
# File 'lib/insika/tools/tts.rb', line 27

def name = "tts"