TTS Voices (#365)

.github/workflows/switch.yml

@@ -89,6 +89,9 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
 
+      - name: Setup Ruby
+        uses: ruby/setup-ruby@v1
+
       - name: Configure AWS credentials
         id: aws-login
         uses: aws-actions/configure-aws-credentials@v4
@@ -116,7 +119,7 @@ jobs:
 
       - name: Export Polly Voices
         run: |
-          components/freeswitch/bin/export_aws_polly_voices components/freeswitch/conf/autoload_configs/polly_voices.xml
+          components/freeswitch/bin/export_tts_voices > components/freeswitch/conf/autoload_configs/tts_voices.xml
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3

.rubocop.yml

@@ -2,6 +2,9 @@ require:
   - rubocop-performance
   - rubocop-rspec
 
+AllCops:
+  NewCops: enable
+
 Style/FrozenStringLiteralComment:
   Enabled: false
 

.tool-versions

@@ -1 +1,2 @@
 terraform 1.5.6
+ruby 3.2.2

components/app/Gemfile

@@ -2,6 +2,8 @@ source "https://rubygems.org"
 
 gem "adhearsion", github: "somleng/adhearsion"
 gem "aws-sdk-lambda"
+gem "aws-sdk-polly"
+gem "blather", github: "adhearsion/blather", branch: "develop"
 gem "faraday"
 gem "http"
 gem "okcomputer"
@@ -11,6 +13,8 @@ gem "sentry-ruby"
 gem "sinatra"
 gem "sinatra-contrib", require: false
 gem "skylight"
+gem "sucker_punch"
+gem "tts_voices", github: "somleng/tts_voices"
 
 group :development, :test do
   gem "rubocop"

components/app/Gemfile.lock

@@ -1,3 +1,14 @@
+GIT
+  remote: https://github.com/adhearsion/blather.git
+  revision: 03f9cda14ed3f93a80504c2538cd514d6b025d04
+  branch: develop
+  specs:
+    blather (2.0.0)
+      activesupport (>= 2.3.11)
+      eventmachine (~> 1.2, >= 1.2.6)
+      niceogiri (~> 1.0)
+      nokogiri (~> 1.8, >= 1.8.3)
+
 GIT
   remote: https://github.com/samnang/reel-rack.git
   revision: 615a9c9219aa00675a4290c4f7245684c701ea0e
@@ -36,20 +47,32 @@ GIT
       thor
       virtus (~> 1.0)
 
+GIT
+  remote: https://github.com/somleng/tts_voices.git
+  revision: afb3b77d74a0bea12b1c8eaa1303e20076a43912
+  specs:
+    tts_voices (0.1.0)
+      aws-sdk-polly
+
 GEM
   remote: https://rubygems.org/
   specs:
-    activesupport (7.0.8)
+    activesupport (7.1.1)
+      base64
+      bigdecimal
       concurrent-ruby (~> 1.0, >= 1.0.2)
+      connection_pool (>= 2.2.5)
+      drb
       i18n (>= 1.6, < 2)
       minitest (>= 5.1)
+      mutex_m
       tzinfo (~> 2.0)
     addressable (2.8.5)
       public_suffix (>= 2.0.2, < 6.0)
     adhearsion-loquacious (1.9.3)
     ast (2.4.2)
     aws-eventstream (1.2.0)
-    aws-partitions (1.835.0)
+    aws-partitions (1.838.0)
     aws-sdk-core (3.185.1)
       aws-eventstream (~> 1, >= 1.0.2)
       aws-partitions (~> 1, >= 1.651.0)
@@ -58,19 +81,17 @@ GEM
     aws-sdk-lambda (1.106.0)
       aws-sdk-core (~> 3, >= 3.184.0)
       aws-sigv4 (~> 1.1)
+    aws-sdk-polly (1.76.0)
+      aws-sdk-core (~> 3, >= 3.184.0)
+      aws-sigv4 (~> 1.1)
     aws-sigv4 (1.6.0)
       aws-eventstream (~> 1, >= 1.0.2)
     axiom-types (0.1.1)
       descendants_tracker (~> 0.0.4)
       ice_nine (~> 0.11.0)
       thread_safe (~> 0.3, >= 0.3.1)
     base64 (0.1.1)
-    blather (2.0.0)
-      activesupport (>= 2.3.11)
-      eventmachine (~> 1.2, >= 1.2.6)
-      niceogiri (~> 1.0)
-      nokogiri (~> 1.8, >= 1.8.3)
-      sucker_punch (~> 2.0)
+    bigdecimal (3.1.4)
     celluloid (0.16.0)
       timers (~> 4.0.0)
     celluloid-io (0.16.2)
@@ -80,6 +101,7 @@ GEM
     coercible (1.0.0)
       descendants_tracker (~> 0.0.1)
     concurrent-ruby (1.2.2)
+    connection_pool (2.4.1)
     countdownlatch (1.0.0)
     crack (0.4.5)
       rexml
@@ -90,6 +112,8 @@ GEM
     docile (1.4.0)
     domain_name (0.5.20190701)
       unf (>= 0.0.5, < 1.0.0)
+    drb (2.1.1)
+      ruby2_keywords
     equalizer (0.0.11)
     eventmachine (1.2.7)
     faraday (2.7.11)
@@ -134,6 +158,7 @@ GEM
     multi_json (1.15.0)
     mustermann (3.0.0)
       ruby2_keywords (~> 0.0.1)
+    mutex_m (0.1.2)
     niceogiri (1.1.2)
       nokogiri (~> 1.5)
     nio4r (2.5.9)
@@ -237,7 +262,7 @@ GEM
     skylight (6.0.1)
       activesupport (>= 5.2.0)
     state_machine (1.2.0)
-    sucker_punch (2.1.2)
+    sucker_punch (3.1.0)
       concurrent-ruby (~> 1.0)
     thor (1.2.2)
     thread_safe (0.3.6)
@@ -274,6 +299,8 @@ PLATFORMS
 DEPENDENCIES
   adhearsion!
   aws-sdk-lambda
+  aws-sdk-polly
+  blather!
   faraday
   http
   okcomputer
@@ -290,6 +317,8 @@ DEPENDENCIES
   sinatra
   sinatra-contrib
   skylight
+  sucker_punch
+  tts_voices!
   twilio-ruby
   vcr
   webmock

components/app/app/call_controllers/call_controller.rb

@@ -64,6 +64,7 @@ def build_call_properties
       api_version: response.api_version,
       to: response.to,
       from: response.from,
+      default_tts_voice: response.default_tts_voice,
       sip_headers: SIPHeaders.new(call_sid: response.call_sid, account_sid: response.account_sid)
     )
   end

components/app/app/jobs/notify_tts_event_job.rb

@@ -0,0 +1,7 @@
+class NotifyTTSEventJob
+  include SuckerPunch::Job
+
+  def perform(client, data)
+    client.notify_tts_event(data)
+  end
+end

components/app/app/models/call_properties.rb

@@ -10,6 +10,7 @@
   :to,
   :from,
   :sip_headers,
+  :default_tts_voice,
   keyword_init: true
 ) do
   def inbound?

components/app/app/models/execute_twiml.rb

@@ -13,6 +13,11 @@ class ExecuteTwiML
   DEFAULT_TWILIO_VOICE = "man".freeze
   DEFAULT_TWILIO_LANGUAGE = "en".freeze
   FINISH_ON_KEY_PATTERN = /\A(?:\d|\*|\#)\z/.freeze
+  BASIC_TTS_MAPPING = {
+    "man" => "Basic.Kal",
+    "woman" => "Basic.Slt"
+  }.freeze
+
   DIAL_CALL_STATUSES = {
     no_answer: "no-answer",
     answer: "completed",
@@ -105,9 +110,17 @@ def execute_say(verb)
     answer unless answered?
 
     attributes = twiml_attributes(verb)
+    tts_voice = resolve_tts_voice(attributes)
+
+    NotifyTTSEventJob.perform_async(
+      call_platform_client,
+      phone_call: call_properties.call_sid,
+      tts_voice: tts_voice.identifier,
+      num_chars: verb.content.length
+    )
 
     twiml_loop(attributes).each do
-      say(say_options(verb.content, attributes))
+      say(say_options(verb.content, tts_voice))
     end
   end
 
@@ -133,7 +146,12 @@ def execute_gather(verb)
       end
 
       nested_verb_attributes = twiml_attributes(nested_verb)
-      content = nested_verb.name == "Say" ? say_options(nested_verb.content, nested_verb_attributes) : nested_verb.content
+      content = if nested_verb.name == "Say"
+        tts_voice = resolve_tts_voice(nested_verb_attributes)
+        say_options(nested_verb.content, tts_voice)
+      else
+        nested_verb.content
+      end
       result.concat(Array.new(twiml_loop(nested_verb_attributes).count, content))
     end
 
@@ -247,14 +265,9 @@ def execute_record(verb)
     )
   end
 
-  def say_options(content, attributes)
-    voice_params = {
-      name: attributes.fetch("voice", DEFAULT_TWILIO_VOICE),
-      language: attributes.fetch("language", DEFAULT_TWILIO_LANGUAGE)
-    }
-
+  def say_options(content, tts_voice)
     ssml = RubySpeech::SSML.draw do
-      voice(voice_params) do
+      voice(name: tts_voice.identifier, language: tts_voice.language) do
         # mod ssml doesn't support non-ascii characters
         # https://github.com/signalwire/freeswitch/issues/1348
         string(content + ".")
@@ -302,4 +315,28 @@ def sip_headers
   def normalize_recording_url(raw_recording_url)
     URL_PATTERN.match(raw_recording_url)[0]
   end
+
+  def resolve_tts_voice(attributes)
+    voice_attribute = attributes["voice"]
+    language_attribute = attributes["language"]
+
+    default_tts_voice = TTSVoices::Voice.find(call_properties.default_tts_voice)
+    voice_attribute = BASIC_TTS_MAPPING.fetch(voice_attribute) if BASIC_TTS_MAPPING.key?(voice_attribute)
+
+    if voice_attribute.blank?
+      tts_voice = resolve_tts_voice_by_language(default_tts_voice, language_attribute)
+      voice_attribute = tts_voice&.identifier
+    end
+
+    TTSVoices::Voice.find(voice_attribute) || default_tts_voice
+  end
+
+  def resolve_tts_voice_by_language(default_tts_voice, language_attribute)
+    return default_tts_voice if language_attribute.blank?
+    return default_tts_voice if default_tts_voice.language.casecmp(language_attribute).zero?
+
+    TTSVoices::Voice.all.find do |voice|
+      voice.provider == default_tts_voice.provider && voice.language.casecmp(language_attribute).zero?
+    end
+  end
 end

components/app/app/models/outbound_call.rb

@@ -29,6 +29,7 @@ def initiate
           api_version: call_params.fetch("api_version"),
           from: call_params.fetch("from"),
           to: call_params.fetch("to"),
+          default_tts_voice: call_params.fetch("default_tts_voice"),
           sip_headers:
         )
       },

components/app/config/initializers/aws_stubs.rb

@@ -28,4 +28,16 @@ def to_h
       }
     }
   }
+
+  Aws.config[:polly] ||= {
+    stub_responses: {
+      describe_voices: {
+        voices: [
+          { gender: "Female", id: "Joanna", language_code: "en-US", supported_engines: ["standard"] },
+          { gender: "Female", id: "Lotte", language_code: "nl-NL", supported_engines: ["neural"] },
+          { gender: "Female", id: "Vitoria", language_code: "pt-BR", supported_engines: %w[neural standard] }
+        ]
+      }
+    }
+  }
 end

components/app/lib/call_platform/client.rb

@@ -15,6 +15,7 @@ class InvalidRequestError < StandardError; end
       :api_version,
       :to,
       :from,
+      :default_tts_voice,
       keyword_init: true
     )
 
@@ -32,6 +33,14 @@ def notify_call_event(params)
       end
     end
 
+    def notify_tts_event(params)
+      response = http_client.post("/services/tts_events", params.to_json)
+
+      unless response.success?
+        Sentry.capture_message("Invalid TTS event", extra: { response_body: response.body })
+      end
+    end
+
     def build_routing_parameters(params)
       make_request("/services/routing_parameters", params: params)
     end
@@ -48,7 +57,8 @@ def create_call(params)
         direction: json_response.fetch("direction"),
         to: json_response.fetch("to"),
         from: json_response.fetch("from"),
-        api_version: json_response.fetch("api_version")
+        api_version: json_response.fetch("api_version"),
+        default_tts_voice: json_response.fetch("default_tts_voice")
       )
     end
 

components/app/lib/call_platform/fake_client.rb

@@ -1,6 +1,7 @@
 module CallPlatform
   class FakeClient < Client
-    def notify_call_event(_params); end
+    def notify_call_event(params); end
+    def notify_tts_event(params); end
 
     def create_call(params)
       validate_gateway_headers(params)
@@ -21,7 +22,8 @@ def create_call(params)
         direction: "inbound",
         to: params.fetch(:to),
         from: params.fetch(:from),
-        api_version: "2010-04-01"
+        api_version: "2010-04-01",
+        default_tts_voice: "Basic.Kal"
       )
     end
 

components/app/spec/call_controllers/gather_spec.rb

@@ -442,7 +442,9 @@
       it "handles nested <Say>" do
         controller = build_controller(
           stub_voice_commands: { ask: build_input_result(nil) },
-          call_properties: { voice_url: "https://www.example.com/gather.xml" }
+          call_properties: {
+            voice_url: "https://www.example.com/gather.xml"
+          }
         )
 
         stub_twiml_request(controller, response: <<~TWIML)
@@ -464,15 +466,15 @@
           outputs.first(3).each do |ssml|
             node = ssml.voice.children.first
             expect(node.content).to eq("Hello World.")
-            expect(node.attributes.fetch("name").value).to eq("woman")
-            expect(node.attributes.fetch("lang").value).to eq("de")
+            expect(node.attributes.fetch("name").value).to eq("Basic.Slt")
+            expect(node.attributes.fetch("lang").value).to eq("en-US")
           end
 
           outputs.last(5).each do |ssml|
             node = ssml.voice.children.first
             expect(node.content).to eq("Foobar.")
-            expect(node.attributes.fetch("name").value).to eq("man")
-            expect(node.attributes.fetch("lang").value).to eq("en")
+            expect(node.attributes.fetch("name").value).to eq("Basic.Kal")
+            expect(node.attributes.fetch("lang").value).to eq("en-US")
           end
         end
       end