Whisper

Dockerfile

@@ -12,13 +12,13 @@
 #   - https://pkgs.org/ - resource for finding needed packages
 #   - Ex: hexpm/elixir:1.12.0-erlang-24.0.1-debian-bullseye-20210902-slim
 #
-ARG BUILDER_IMAGE="hexpm/elixir:1.12.0-erlang-24.0.1-debian-bullseye-20210902-slim"
+ARG BUILDER_IMAGE="hexpm/elixir:1.14.0-erlang-24.0.1-debian-bullseye-20210902-slim"
 ARG RUNNER_IMAGE="debian:bullseye-20210902-slim"
 
 FROM ${BUILDER_IMAGE} as builder
 
 # install build dependencies
-RUN apt-get update -y && apt-get install -y build-essential git \
+RUN apt-get update -y && apt-get install -y build-essential git curl ffmpeg \
     && apt-get clean && rm -f /var/lib/apt/lists/*_*
 
 # prepare build dir
@@ -30,6 +30,7 @@ RUN mix local.hex --force && \
 
 # set build ENV
 ENV MIX_ENV="prod"
+ENV BUMBLEBEE_CACHE_DIR="/app/.bumblebee"
 
 # install mix dependencies
 COPY mix.exs mix.lock ./
@@ -57,6 +58,7 @@ COPY assets assets
 RUN mix assets.deploy
 
 RUN mix compile
+RUN mix run -e 'LiveBeats.Application.load_serving()' --no-start
 
 # Changes to config/runtime.exs don't require recompiling the code
 COPY config/runtime.exs config/
@@ -68,7 +70,7 @@ RUN mix release
 # the compiled release and other runtime necessities
 FROM ${RUNNER_IMAGE}
 
-RUN apt-get update -y && apt-get install -y libstdc++6 openssl libncurses5 locales \
+RUN apt-get update -y && apt-get install -y libstdc++6 openssl libncurses5 locales curl ffmpeg \
   && apt-get clean && rm -f /var/lib/apt/lists/*_*
 
 # Set the locale
@@ -80,9 +82,11 @@ ENV LC_ALL en_US.UTF-8
 
 WORKDIR "/app"
 RUN chown nobody /app
+ENV BUMBLEBEE_CACHE_DIR="/app/.bumblebee"
 
 # Only copy the final release from the build stage
 COPY --from=builder --chown=nobody:root /app/_build/prod/rel/live_beats ./
+COPY --from=builder --chown=nobody:root /app/.bumblebee/ ./.bumblebee
 
 USER nobody
 

config/config.exs

@@ -34,7 +34,7 @@ config :esbuild,
 
 # Configure tailwind (the version is required)
 config :tailwind,
-  version: "3.1.8",
+  version: "3.2.7",
   default: [
     args: ~w(
       --config=tailwind.config.js

config/dev.exs

@@ -2,7 +2,7 @@ import Config
 
 config :live_beats, :files,
   uploads_dir: Path.expand("../priv/uploads", __DIR__),
-  host: [scheme: "http", host: "localhost", port: 4000],
+  host: [scheme: "http", host: "localhost", port: 4001],
   server_ip: "127.0.0.1",
   hostname: "localhost",
   transport_opts: []

fly.toml

@@ -1,45 +1,46 @@
-app = "livebeats"
+# fly.toml app configuration file generated for livebeats on 2023-05-19T22:42:40-04:00
+#
+# See https://fly.io/docs/reference/configuration/ for information about how to use this file.
+#
 
+app = "livebeats"
+primary_region = "ord"
 kill_signal = "SIGTERM"
-kill_timeout = 5
-processes = []
+kill_timeout = "5s"
+
+[experimental]
+  auto_rollback = true
 
 [deploy]
   release_command = "/app/bin/migrate"
 
 [env]
+  BUMBLEBEE_CACHE_DIR = "/app/.bumblebee"
   PHX_HOST = "livebeats.fly.dev"
 
 [mounts]
   source="data"
   destination="/app/uploads"
 
-[experimental]
-  allowed_public_ports = []
-  auto_rollback = true
-
 [[services]]
-  http_checks = []
+  protocol = "tcp"
   internal_port = 4000
   processes = ["app"]
-  protocol = "tcp"
-  script_checks = []
 
+  [[services.ports]]
+    port = 80
+    handlers = ["http"]
+
+  [[services.ports]]
+    port = 443
+    handlers = ["tls", "http"]
   [services.concurrency]
+    type = "connections"
     hard_limit = 2500
     soft_limit = 2000
-    type = "connections"
-
-  [[services.ports]]
-    handlers = ["http"]
-    port = 80
-
-  [[services.ports]]
-    handlers = ["tls", "http"]
-    port = 443
 
   [[services.tcp_checks]]
-    grace_period = "20s"  # allow some time for startup
     interval = "15s"
+    timeout = "2s"
+    grace_period = "20s"
     restart_limit = 0
-    timeout = "2s"

lib/live_beats/application.ex

@@ -5,32 +5,46 @@ defmodule LiveBeats.Application do
 
   use Application
 
+  def load_serving do
+    {:ok, whisper} = Bumblebee.load_model({:hf, "openai/whisper-tiny"})
+    {:ok, featurizer} = Bumblebee.load_featurizer({:hf, "openai/whisper-tiny"})
+    {:ok, tokenizer} = Bumblebee.load_tokenizer({:hf, "openai/whisper-tiny"})
+
+    Bumblebee.Audio.speech_to_text(whisper, featurizer, tokenizer,
+      compile: [batch_size: 1],
+      max_new_tokens: 100,
+      defn_options: [compiler: EXLA]
+    )
+  end
+
   @impl true
   def start(_type, _args) do
     LiveBeats.MediaLibrary.attach()
     topologies = Application.get_env(:libcluster, :topologies) || []
 
-    children = [
-      {Cluster.Supervisor, [topologies, [name: LiveBeats.ClusterSupervisor]]},
-      {Task.Supervisor, name: LiveBeats.TaskSupervisor},
-      # Start the Ecto repository
-      LiveBeats.Repo,
-      LiveBeats.ReplicaRepo,
-      # Start the Telemetry supervisor
-      LiveBeatsWeb.Telemetry,
-      # Start the PubSub system
-      {Phoenix.PubSub, name: LiveBeats.PubSub},
-      # start presence
-      LiveBeatsWeb.Presence,
-      {Finch, name: LiveBeats.Finch},
-      # Start the Endpoint (http/https)
-      LiveBeatsWeb.Endpoint,
-      # Expire songs every six hours
-      {LiveBeats.SongsCleaner, interval: {3600 * 6, :second}}
-
-      # Start a worker by calling: LiveBeats.Worker.start_link(arg)
-      # {LiveBeats.Worker, arg}
-    ]
+    children =
+      [
+        {Nx.Serving, name: WhisperServing, serving: load_serving()},
+        {Cluster.Supervisor, [topologies, [name: LiveBeats.ClusterSupervisor]]},
+        {Task.Supervisor, name: LiveBeats.TaskSupervisor},
+        {Task.Supervisor, name: Fly.Machine.TaskSupervisor},
+        # Start the Ecto repository
+        LiveBeats.Repo,
+        LiveBeats.ReplicaRepo,
+        # Start the Telemetry supervisor
+        LiveBeatsWeb.Telemetry,
+        # Start the PubSub system
+        {Phoenix.PubSub, name: LiveBeats.PubSub},
+        # start presence
+        LiveBeatsWeb.Presence,
+        {Finch, name: LiveBeats.Finch},
+        # Start the Endpoint (http/https)
+        LiveBeatsWeb.Endpoint,
+        # Expire songs every six hours
+        {LiveBeats.SongsCleaner, interval: {3600 * 6, :second}}
+        # Start a worker by calling: LiveBeats.Worker.start_link(arg)
+        # {LiveBeats.Worker, arg}
+      ]
 
     # See https://hexdocs.pm/elixir/Supervisor.html
     # for other strategies and supported options

lib/live_beats/audio.ex

@@ -0,0 +1,17 @@
+defmodule LiveBeats.Audio do
+  def speech_to_text(path, chunk_time, func) do
+    {:ok, stat} = LiveBeats.MP3Stat.parse(path)
+
+    0..stat.duration//chunk_time
+    |> Task.async_stream(
+      fn ss ->
+        args = ~w(-ac 1 -ar 16k -f f32le -ss #{ss} -t #{chunk_time} -v quiet -)
+        {data, 0} = System.cmd("ffmpeg", ["-i", path] ++ args)
+        {ss, Nx.Serving.batched_run(WhisperServing, Nx.from_binary(data, :f32))}
+      end,
+      max_concurrency: 2,
+      timeout: :infinity
+    )
+    |> Enum.map(fn {:ok, {ss, %{results: [%{text: text}]}}} -> func.(ss, text) end)
+  end
+end

lib/live_beats/media_library.ex

@@ -52,13 +52,7 @@ defmodule LiveBeats.MediaLibrary do
     user.id == song.user_id
   end
 
-  def play_song(%Song{id: id}) do
-    play_song(id)
-  end
-
-  def play_song(id) do
-    song = get_song!(id)
-
+  def play_song(%Song{} = song) do
     played_at =
       cond do
         playing?(song) ->
@@ -97,6 +91,12 @@ defmodule LiveBeats.MediaLibrary do
     new_song
   end
 
+  def play_song(id) do
+    id
+    |> get_song!()
+    |> play_song()
+  end
+
   def pause_song(%Song{} = song) do
     now = DateTime.truncate(DateTime.utc_now(), :second)
     set = [status: :paused, paused_at: now]
@@ -211,6 +211,7 @@ defmodule LiveBeats.MediaLibrary do
           |> Enum.filter(&match?({{:song, _ref}, _}, &1))
           |> Enum.map(fn {{:song, ref}, song} ->
             consume_file.(ref, fn tmp_path -> store_mp3(song, tmp_path) end)
+            async_transcribe(song, user)
 
             {ref, song}
           end)
@@ -231,6 +232,22 @@ defmodule LiveBeats.MediaLibrary do
     end
   end
 
+  defp async_transcribe(%Song{} = song, %Accounts.User{} = user) do
+    Task.Supervisor.start_child(LiveBeats.TaskSupervisor, fn ->
+      segments =
+        LiveBeats.Audio.speech_to_text(song.mp3_filepath, 20, fn ss, text ->
+          segment = %Song.TranscriptSegment{ss: ss, text: text}
+          broadcast!(user.id, {segment, song.id})
+
+          segment
+        end)
+
+      Repo.update_all(from(s in Song, where: s.id == ^song.id),
+        set: [transcript_segments: segments]
+      )
+    end)
+  end
+
   defp broadcast_imported(%Accounts.User{} = user, songs) do
     songs = Enum.map(songs, fn {_ref, song} -> song end)
     broadcast!(user.id, %Events.SongsImported{user_id: user.id, songs: songs})

lib/live_beats/media_library/song.ex

@@ -25,6 +25,11 @@ defmodule LiveBeats.MediaLibrary.Song do
     belongs_to :user, Accounts.User
     belongs_to :genre, LiveBeats.MediaLibrary.Genre
 
+    embeds_many :transcript_segments, TranscriptSegment do
+      field :ss, :integer
+      field :text, :string
+    end
+
     timestamps()
   end
 

lib/live_beats_web/controllers/oauth_callback_controller.ex

@@ -15,7 +15,7 @@ defmodule LiveBeatsWeb.OAuthCallbackController do
       |> LiveBeatsWeb.UserAuth.log_in_user(user)
     else
       {:error, %Ecto.Changeset{} = changeset} ->
-        Logger.debug("failed GitHub insert #{inspect(changeset.errors)}")
+        Logger.info("failed GitHub insert #{inspect(changeset.errors)}")
 
         conn
         |> put_flash(
@@ -25,7 +25,7 @@ defmodule LiveBeatsWeb.OAuthCallbackController do
         |> redirect(to: "/")
 
       {:error, reason} ->
-        Logger.debug("failed GitHub exchange #{inspect(reason)}")
+        Logger.info("failed GitHub exchange #{inspect(reason)}")
 
         conn
         |> put_flash(:error, "We were unable to contact GitHub. Please try again later")

lib/live_beats_web/live/profile_live.ex

@@ -60,6 +60,13 @@ defmodule LiveBeatsWeb.ProfileLive do
       total_count={@presences_count}
     />
 
+    <div id={"trascript-#{@active_song_id}"} phx-update="stream" class="mt-10 px-6">
+      <div :for={{id, segment} <- @streams.transcript_segments} id={id}>
+        <span class="text-gray-400">[<%= LiveBeats.MP3Stat.to_mmss(segment.ss) %>]</span>
+        <%= segment.text %>
+      </div>
+    </div>
+
     <div id="dialogs" phx-update="stream">
       <%= for {_id, song} <- if(@owns_profile?, do: @streams.songs, else: []), id = "delete-modal-#{song.id}" do %>
         <.modal
@@ -151,6 +158,7 @@ defmodule LiveBeatsWeb.ProfileLive do
     end
 
     active_song = MediaLibrary.get_current_active_song(profile)
+    segments = if active_song, do: active_song.transcript_segments, else: []
 
     songs = MediaLibrary.list_profile_songs(profile, 50)
 
@@ -164,6 +172,7 @@ defmodule LiveBeatsWeb.ProfileLive do
         songs_count: Enum.count(songs)
       )
       |> stream(:songs, songs)
+      |> stream(:transcript_segments, segments, dom_id: &"ss-#{&1.ss}")
       |> assign_presences()
 
     {:ok, socket, temporary_assigns: [presences: %{}]}
@@ -255,12 +264,20 @@ defmodule LiveBeatsWeb.ProfileLive do
     {:noreply, pause_song(socket, song)}
   end
 
+  def handle_info({MediaLibrary, {%MediaLibrary.Song.TranscriptSegment{} = seg, song_id}}, socket) do
+    if socket.assigns.active_song_id == song_id do
+      {:noreply, stream_insert(socket, :transcript_segments, seg)}
+    else
+      {:noreply, socket}
+    end
+  end
+
   def handle_info({MediaLibrary, %MediaLibrary.Events.SongsImported{songs: songs}}, socket) do
     %{current_user: current_user, active_song_id: active_song_id} = socket.assigns
     first = hd(songs)
 
     if !active_song_id && MediaLibrary.can_control_playback?(current_user, first) do
-      MediaLibrary.play_song(first.id)
+      MediaLibrary.play_song(first)
     end
 
     {:noreply,

mix.exs

@@ -53,8 +53,11 @@ defmodule LiveBeats.MixProject do
       {:mint, "~> 1.0"},
       {:heroicons, "~> 0.2.2"},
       {:castore, "~> 0.1.13"},
-      {:tailwind, "~> 0.1"},
-      {:libcluster, "~> 3.3.1"}
+      {:tailwind, "~> 0.2.0"},
+      {:libcluster, "~> 3.3.1"},
+      {:bumblebee, github: "elixir-nx/bumblebee"},
+      {:exla, ">= 0.0.0"},
+      {:req, "~> 0.3.7"}
     ]
   end
 

mix.lock

@@ -45,9 +45,10 @@
   "postgrex": {:hex, :postgrex, "0.16.5", "fcc4035cc90e23933c5d69a9cd686e329469446ef7abba2cf70f08e2c4b69810", [:mix], [{:connection, "~> 1.1", [hex: :connection, repo: "hexpm", optional: false]}, {:db_connection, "~> 2.1", [hex: :db_connection, repo: "hexpm", optional: false]}, {:decimal, "~> 1.5 or ~> 2.0", [hex: :decimal, repo: "hexpm", optional: false]}, {:jason, "~> 1.0", [hex: :jason, repo: "hexpm", optional: true]}, {:table, "~> 0.1.0", [hex: :table, repo: "hexpm", optional: true]}], "hexpm", "edead639dc6e882618c01d8fc891214c481ab9a3788dfe38dd5e37fd1d5fb2e8"},
   "progress_bar": {:hex, :progress_bar, "2.0.1", "7b40200112ae533d5adceb80ff75fbe66dc753bca5f6c55c073bfc122d71896d", [:mix], [{:decimal, "~> 2.0", [hex: :decimal, repo: "hexpm", optional: false]}], "hexpm", "2519eb58a2f149a3a094e729378256d8cb6d96a259ec94841bd69fdc71f18f87"},
   "ranch": {:hex, :ranch, "1.8.0", "8c7a100a139fd57f17327b6413e4167ac559fbc04ca7448e9be9057311597a1d", [:make, :rebar3], [], "hexpm", "49fbcfd3682fab1f5d109351b61257676da1a2fdbe295904176d5e521a2ddfe5"},
+  "req": {:hex, :req, "0.3.7", "e4ea5d73e3f434c0a15601bb85330ffd0e57860c098283e98c28d21172a1f749", [:mix], [{:brotli, "~> 0.3.1", [hex: :brotli, repo: "hexpm", optional: true]}, {:ezstd, "~> 1.0", [hex: :ezstd, repo: "hexpm", optional: true]}, {:finch, "~> 0.9", [hex: :finch, repo: "hexpm", optional: false]}, {:jason, "~> 1.0", [hex: :jason, repo: "hexpm", optional: false]}, {:mime, "~> 1.6 or ~> 2.0", [hex: :mime, repo: "hexpm", optional: false]}, {:nimble_csv, "~> 1.0", [hex: :nimble_csv, repo: "hexpm", optional: true]}, {:plug, "~> 1.0", [hex: :plug, repo: "hexpm", optional: true]}], "hexpm", "a7d3c0bec7d2d23198ef12676d2c950bec258308c6a5123eb98465030205f39c"},
   "rustler_precompiled": {:hex, :rustler_precompiled, "0.6.1", "160b545bce8bf9a3f1b436b2c10f53574036a0db628e40f393328cbbe593602f", [:mix], [{:castore, "~> 0.1 or ~> 1.0", [hex: :castore, repo: "hexpm", optional: false]}, {:rustler, "~> 0.23", [hex: :rustler, repo: "hexpm", optional: true]}], "hexpm", "0dd269fa261c4e3df290b12031c575fff07a542749f7b0e8b744d72d66c43600"},
   "swoosh": {:hex, :swoosh, "1.8.2", "af9a22ab2c0d20b266f61acca737fa11a121902de9466a39e91bacdce012101c", [:mix], [{:cowboy, "~> 1.1 or ~> 2.4", [hex: :cowboy, repo: "hexpm", optional: true]}, {:ex_aws, "~> 2.1", [hex: :ex_aws, repo: "hexpm", optional: true]}, {:finch, "~> 0.6", [hex: :finch, repo: "hexpm", optional: true]}, {:gen_smtp, "~> 0.13 or ~> 1.0", [hex: :gen_smtp, repo: "hexpm", optional: true]}, {:hackney, "~> 1.9", [hex: :hackney, repo: "hexpm", optional: true]}, {:jason, "~> 1.0", [hex: :jason, repo: "hexpm", optional: false]}, {:mail, "~> 0.2", [hex: :mail, repo: "hexpm", optional: true]}, {:mime, "~> 1.1 or ~> 2.0", [hex: :mime, repo: "hexpm", optional: false]}, {:plug_cowboy, ">= 1.0.0", [hex: :plug_cowboy, repo: "hexpm", optional: true]}, {:telemetry, "~> 0.4.2 or ~> 1.0", [hex: :telemetry, repo: "hexpm", optional: false]}], "hexpm", "d058ba750eafadb6c09a84a352c14c5d1eeeda6e84945fcc95785b7f3067b7db"},
-  "tailwind": {:hex, :tailwind, "0.1.9", "25ba09d42f7bfabe170eb67683a76d6ec2061952dc9bd263a52a99ba3d24bd4d", [:mix], [{:castore, ">= 0.0.0", [hex: :castore, repo: "hexpm", optional: false]}], "hexpm", "9213f87709c458aaec313bb5f2df2b4d2cedc2b630e4ae821bf3c54c47a56d0b"},
+  "tailwind": {:hex, :tailwind, "0.2.0", "95f9e4a32020c5bec480f1d6a43a49ac8030b13183127b577605f506d6e13a66", [:mix], [{:castore, ">= 0.0.0", [hex: :castore, repo: "hexpm", optional: false]}], "hexpm", "385e939fcd7fe4654be5130b187e358aaabade385513f9d200ffecdbb9552a9e"},
   "telemetry": {:hex, :telemetry, "1.2.1", "68fdfe8d8f05a8428483a97d7aab2f268aaff24b49e0f599faa091f1d4e7f61c", [:rebar3], [], "hexpm", "dad9ce9d8effc621708f99eac538ef1cbe05d6a874dd741de2e689c47feafed5"},
   "telemetry_metrics": {:hex, :telemetry_metrics, "0.6.1", "315d9163a1d4660aedc3fee73f33f1d355dcc76c5c3ab3d59e76e3edf80eef1f", [:mix], [{:telemetry, "~> 0.4 or ~> 1.0", [hex: :telemetry, repo: "hexpm", optional: false]}], "hexpm", "7be9e0871c41732c233be71e4be11b96e56177bf15dde64a8ac9ce72ac9834c6"},
   "telemetry_poller": {:hex, :telemetry_poller, "1.0.0", "db91bb424e07f2bb6e73926fcafbfcbcb295f0193e0a00e825e589a0a47e8453", [:rebar3], [{:telemetry, "~> 1.0", [hex: :telemetry, repo: "hexpm", optional: false]}], "hexpm", "b3a24eafd66c3f42da30fc3ca7dda1e9d546c12250a2d60d7b81d264fbec4f6e"},

priv/repo/migrations/20230314150807_add_transcripts_to_songs.exs

@@ -0,0 +1,9 @@
+defmodule LiveBeats.Repo.Migrations.AddTranscriptsToSongs do
+  use Ecto.Migration
+
+  def change do
+    alter table(:songs) do
+      add :transcript_segments, {:array, :map}, null: false, default: []
+    end
+  end
+end