Upload whisper_tests.livemd

public-apps/whisper_tests.livemd

@@ -0,0 +1,78 @@
+<!-- livebook:{"app_settings":{"access_type":"public","auto_shutdown_ms":3600000,"multi_session":true,"output_type":"rich","slug":"whisper-tests"}} -->
+
+# Whisper Tests
+
+```elixir
+Mix.install(
+  [
+    {:bumblebee, github: "elixir-nx/bumblebee"},
+    {:kino_bumblebee, github: "livebook-dev/kino_bumblebee"},
+    {:kino, github: "livebook-dev/kino", override: true},
+    {:exla, "~> 0.6.1"}
+  ],
+  config: [nx: [default_backend: EXLA.Backend]]
+)
+```
+
+## Section
+
+```elixir
+language = Kino.Input.select("Select Language", en: "English", tr: "Turkish")
+```
+
+<!-- livebook:{"reevaluate_automatically":true} -->
+
+```elixir
+{:ok, model_info} = Bumblebee.load_model({:hf, "openai/whisper-base"})
+{:ok, featurizer} = Bumblebee.load_featurizer({:hf, "openai/whisper-base"})
+{:ok, tokenizer} = Bumblebee.load_tokenizer({:hf, "openai/whisper-base"})
+
+{:ok, generation_config} =
+  Bumblebee.load_generation_config({:hf, "openai/whisper-base"})
+
+generation_config = Bumblebee.configure(generation_config, max_new_tokens: 100)
+
+# Kino.listen(language_form, fn event ->
+#   lang = Atom.to_string(event.data.name)
+#   #  IO.puts(event.data.name)
+
+# end)
+
+lang = Kino.Input.read(language)
+
+serving =
+  Bumblebee.Audio.speech_to_text_whisper(model_info, featurizer, tokenizer, generation_config,
+    compile: [batch_size: 30],
+    defn_options: [compiler: EXLA],
+    chunk_num_seconds: 30,
+    language: Atom.to_string(lang)
+  )
+```
+
+<!-- livebook:{"reevaluate_automatically":true} -->
+
+```elixir
+audio_input = Kino.Input.audio("Audio", sampling_rate: featurizer.sampling_rate)
+
+form = Kino.Control.form([audio: audio_input], submit: "Run")
+frame = Kino.Frame.new()
+
+Kino.listen(form, fn %{data: %{audio: audio}} ->
+  if audio do
+    Kino.Frame.render(frame, Kino.Text.new("Running..."))
+
+    audio =
+      audio.data
+      |> Nx.from_binary(:f32)
+      |> Nx.reshape({:auto, audio.num_channels})
+      |> Nx.mean(axes: [1])
+
+    output = Nx.Serving.run(serving, audio)
+    # %{chunks: [%{text: generated_text}]}
+    text = output.chunks |> Enum.map_join(& &1.text) |> String.trim()
+    Kino.Frame.render(frame, Kino.Text.new(text))
+  end
+end)
+
+Kino.Layout.grid([form, frame], boxed: true, gap: 16)
+```