typecast-ruby 0.1.6 → 0.1.7

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: f65c6725ca10d449b9ba688231a8ad96a7042197efe9fdba188b84a55fa1d232
-  data.tar.gz: 9995e18d6b6364c9e9906216d643759ab0570b0867228cb49951a0e7259ccb2e
+  metadata.gz: f3d9076c37ff2cfbc76912eb4ac30f63edfbcd7ab0dc40ad403a63f7532a29c7
+  data.tar.gz: 81459d58895a8d94fb11b8dfed40b69b3ad7c8b7cab2c6d3df4d651d797c0cd0
 SHA512:
-  metadata.gz: e724bac05758d13d390cb1865eb522c57b55cdd6ae976d778d28643b06db35f390e3151040a4f611138977a13d6f07448ec35f39cb86f13ce7376915b2820c50
-  data.tar.gz: 75001ca132eed41352314a431fc34d184f5d3f8c931f3cbce6b5809ca9f90214da108c6f97c7bf48d048744a325ee15783c2bc9d75faef053a74c3fcf93c0dc2
+  metadata.gz: 7819e4ed4127aa8277840cebe895a27ccf58e34e07be121b51ff48f236bf85b353b4d234a564a99ef9bac13fcae6e2c2fc1b9f3ec2208d6c166fad5ac754058c
+  data.tar.gz: de36d1e5f96ca440863658cd2eba53bff83f3452815c04d2d4b842dc4a00b81528c1bebc3c951b297fb73ebd6c60e6118d9c12755b39c1bab080a541e90fc389

data/lib/typecast/client.rb

@@ -29,14 +29,24 @@ module Typecast
       Models::TTSResponse.new(
         audio_data: response.body,
         duration: response["X-Audio-Duration"].to_f,
-        format: response["Content-Type"].to_s.include?("mp3") || response["Content-Type"].to_s.include?("mpeg") ? Models::AUDIO_MP3 : Models::AUDIO_WAV
+        format: response["Content-Type"].to_s.downcase.include?("mp3") || response["Content-Type"].to_s.downcase.include?("mpeg") ? Models::AUDIO_MP3 : Models::AUDIO_WAV
       )
     end
 
     def compose_speech
-      SpeechComposer.new(method(:text_to_speech))
+      SpeechComposer.new(method(:compose_text_to_speech))
     end
 
+    def compose_text_to_speech(segments)
+      response = request_json(:post, "/v1/text-to-speech/compose", segments: segments)
+      Models::TTSResponse.new(
+        audio_data: response.body,
+        duration: response["X-Audio-Duration"].to_f,
+        format: response["Content-Type"].to_s.downcase.include?("mp3") || response["Content-Type"].to_s.downcase.include?("mpeg") ? Models::AUDIO_MP3 : Models::AUDIO_WAV
+      )
+    end
+    private :compose_text_to_speech
+
     # Browse available API voices at https://typecast.ai/developers/api/voices.
     def generate_to_file(path, text:, voice_id:, model: Models::TTS_MODEL_V30, language: nil, prompt: nil, output: nil, seed: nil)
       request = Models::TTSRequest.new(

data/lib/typecast/composer.rb

@@ -1,5 +1,3 @@
-require "stringio"
-
 require "typecast/models"
 
 module Typecast
@@ -13,10 +11,12 @@ module Typecast
     last_index = 0
     text.to_s.scan(PAUSE_TOKEN) do |match|
       match_data = Regexp.last_match
+      seconds = match[0].to_f
+      next unless seconds.finite? && seconds.positive?
       if match_data.begin(0) > last_index
         parts << TextPart.new(kind: "text", text: text[last_index...match_data.begin(0)])
       end
-      parts << PausePart.new(kind: "pause", seconds: match[0].to_f)
+      parts << PausePart.new(kind: "pause", seconds: seconds)
       last_index = match_data.end(0)
     end
     if last_index < text.length
@@ -26,8 +26,8 @@ module Typecast
   end
 
   class SpeechComposer
-    def initialize(text_to_speech)
-      @text_to_speech = text_to_speech
+    def initialize(compose)
+      @compose = compose
       @defaults = {}
       @parts = []
     end
@@ -78,39 +78,22 @@ module Typecast
         raise ArgumentError, "at least one speech segment is required"
       end
 
-      output_format = @defaults.dig(:output, :audio_format) || Models::AUDIO_WAV
+      formats = plan.each_with_object([]) do |part, values|
+        format = part.is_a?(Hash) && part[:kind] == "speech" ? part.dig(:settings, :output, :audio_format) : nil
+        values << format if format
+      end.uniq
+      raise ArgumentError, "composed speech segments must use one audio format" if formats.length > 1
+
+      output_format = formats.first || Models::AUDIO_WAV
       unless [Models::AUDIO_WAV, Models::AUDIO_MP3].include?(output_format)
         raise ArgumentError, "unsupported composed speech output format: #{output_format}"
       end
 
-      wav_spec = nil
-      output_samples = []
-      plan.each do |part|
-        if part.is_a?(PausePart)
-          raise ArgumentError, "pause cannot be the first composed part" if wav_spec.nil?
-
-          output_samples.concat(Array.new(seconds_to_samples(part.seconds, wav_spec[:sample_rate]), 0))
-          next
-        end
-
-        response = @text_to_speech.call(request_from_settings(part[:text], part[:settings]))
-        wav = parse_wav(response.audio_data)
-        if wav_spec && wav[:spec] != wav_spec
-          raise ArgumentError, "all composed WAV segments must use the same PCM format"
-        end
-
-        wav_spec = wav[:spec]
-        output_samples.concat(trim_silence(wav[:samples]))
+      segments = plan.map do |part|
+        part.is_a?(PausePart) ? { type: "pause", duration_seconds: part.seconds } :
+          { type: "tts", **request_from_settings(part[:text], part[:settings], output_format).to_h }
       end
-
-      wav_data = encode_wav(output_samples, wav_spec)
-      raise ArgumentError, "ffmpeg is required to encode composed speech as mp3" if output_format == Models::AUDIO_MP3
-
-      Models::TTSResponse.new(
-        audio_data: wav_data,
-        duration: output_samples.length.to_f / wav_spec[:sample_rate],
-        format: Models::AUDIO_WAV
-      )
+      @compose.call(segments)
     end
 
     private
@@ -169,8 +152,8 @@ module Typecast
       output
     end
 
-    def request_from_settings(text, settings)
-      output = merge_output(settings[:output], audio_format: Models::AUDIO_WAV)
+    def request_from_settings(text, settings, output_format)
+      output = merge_output(settings[:output], audio_format: output_format)
       Models::TTSRequest.new(
         voice_id: settings[:voice_id],
         text: text,
@@ -182,68 +165,5 @@ module Typecast
       )
     end
 
-    def parse_wav(data)
-      io = StringIO.new(data)
-      raise ArgumentError, "unsupported WAV data" unless io.read(4) == "RIFF"
-
-      io.read(4)
-      raise ArgumentError, "unsupported WAV data" unless io.read(4) == "WAVE"
-
-      spec = nil
-      samples = nil
-      until io.eof?
-        chunk_id = io.read(4)
-        break if chunk_id.nil? || chunk_id.bytesize < 4
-
-        chunk_size_bytes = io.read(4)
-        raise ArgumentError, "unsupported WAV data" if chunk_size_bytes.nil? || chunk_size_bytes.bytesize < 4
-
-        chunk_size = chunk_size_bytes.unpack1("V")
-        chunk_data = io.read(chunk_size)
-        io.read(1) if chunk_size.odd?
-        raise ArgumentError, "unsupported WAV data" if chunk_data.nil? || chunk_data.bytesize < chunk_size
-
-        case chunk_id
-        when "fmt "
-          audio_format, channels, sample_rate, _byte_rate, _block_align, bits_per_sample = chunk_data.unpack("vvVVvv")
-          if audio_format != 1 || channels != 1 || bits_per_sample != 16
-            raise ArgumentError, "only mono 16-bit PCM WAV is supported for composed speech"
-          end
-          spec = { sample_rate: sample_rate, channels: channels, bits_per_sample: bits_per_sample }
-        when "data"
-          samples = chunk_data.unpack("s<*")
-        end
-      end
-
-      raise ArgumentError, "unsupported WAV data" if spec.nil? || samples.nil?
-
-      { spec: spec, samples: samples }
-    end
-
-    def encode_wav(samples, spec)
-      payload = samples.pack("s<*")
-      [
-        "RIFF",
-        [36 + payload.bytesize].pack("V"),
-        "WAVE",
-        "fmt ",
-        [16, 1, spec[:channels], spec[:sample_rate], spec[:sample_rate] * spec[:channels] * 2, spec[:channels] * 2, spec[:bits_per_sample]].pack("VvvVVvv"),
-        "data",
-        [payload.bytesize].pack("V"),
-        payload
-      ].join
-    end
-
-    def trim_silence(samples)
-      start_index = 0
-      end_index = samples.length
-      start_index += 1 while start_index < end_index && samples[start_index].abs <= 0
-      end_index -= 1 while end_index > start_index && samples[end_index - 1].abs <= 0
-      samples[start_index...end_index] || []
-    end
-
-    def seconds_to_samples(seconds, sample_rate)
-      (seconds * sample_rate).round
-    end
   end
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: typecast-ruby
 version: !ruby/object:Gem::Version
-  version: 0.1.6
+  version: 0.1.7
 platform: ruby
 authors:
 - Neosapience
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2026-07-02 00:00:00.000000000 Z
+date: 2026-07-16 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: minitest