mejora dictado

whisper_live/lib/whisper_live/audio_buffer.ex

@@ -1,27 +1,106 @@
 defmodule WhisperLive.AudioBuffer do
-  use GenServer
+    use GenServer
 
-  ## API
+    # API
 
-  def start_link(ref), do: GenServer.start_link(__MODULE__, [], name: via(ref))
+    def start_link(ref), do: GenServer.start_link(__MODULE__, [], name: via(ref))
 
-  def append(ref, chunk), do: GenServer.cast(via(ref), {:append, chunk})
+    def append(ref, chunk), do: GenServer.cast(via(ref), {:append, chunk})
 
-  def get_all(ref), do: GenServer.call(via(ref), :get_all)
+    def get_all(ref), do: GenServer.call(via(ref), :get_all)
 
-  def clear(ref), do: GenServer.call(via(ref), :clear)
+    def get_tiny(ref), do: GenServer.call(via(ref), :get_tiny)       
 
-  def stop(ref), do: GenServer.stop(via(ref))
+    def clear_tiny(ref), do: GenServer.call(via(ref), :clear_tiny)     
 
-  defp via(ref), do: {:via, Registry, {WhisperLive.AudioRegistry, ref}}
+    def clear(ref), do: GenServer.call(via(ref), :clear)
 
-  ## Callbacks
+    def stop(ref), do: GenServer.stop(via(ref))
 
-  def init(_), do: {:ok, []}
+    def pop_chunk_with_overlap(ref, keep_ms \\ 1000), do: GenServer.call(via(ref), {:pop_chunk_with_overlap, keep_ms})
 
-  def handle_cast({:append, chunk}, state), do: {:noreply, [chunk | state]}
+    defp via(ref), do: {:via, Registry, {WhisperLive.AudioRegistry, ref}}
 
-  def handle_call(:get_all, _from, state), do: {:reply, Enum.reverse(state), state}
+    # Callbacks
 
-  def handle_call(:clear, _from, _state), do: {:reply, :ok, []}
+    def init(_), do: {:ok, %{tiny: [], full: []}}
+
+    def handle_cast({:append, chunk}, state) do
+        {:noreply, %{state | tiny: [chunk | state.tiny], full: [chunk | state.full]}}
+    end
+
+    @impl true
+    def handle_call({:pop_chunk_with_overlap, keep_ms}, _from, state) do
+        bytes_per_ms = div(state.rate * 2, 1000)
+        total = state.last_overlap <> state.buffer
+
+        total_size = byte_size(total)
+        keep_bytes = keep_ms * bytes_per_ms
+        overlap_bytes = min(keep_bytes, total_size)
+
+        {to_send, to_keep} = split_bytes(total, overlap_bytes)
+
+        new_state = %{state | buffer: <<>>, last_overlap: to_keep}
+        {:reply, {to_send, state.rate}, new_state}
+    end
+
+    defp split_bytes(binary, keep_bytes) do
+        total = byte_size(binary)
+        send_bytes = max(total - keep_bytes, 0)
+        <<send::binary-size(send_bytes), keep::binary>> = binary
+        {send, keep}
+    end
+
+    def handle_call(:get_all, _from, state), do: {:reply, Enum.reverse(state.full), state}
+    def handle_call(:get_tiny, _from, state), do: {:reply, Enum.reverse(state.tiny), state}
+    def handle_call(:clear_tiny, _from, state), do: {:reply, :ok, %{state | tiny: []}}
+    def handle_call(:clear, _from, _state), do: {:reply, :ok, %{tiny: [], full: []}}
 end
+
+# defmodule WhisperLive.AudioBuffer do
+#     use GenServer
+
+#     def start_link(ref), do: GenServer.start_link(__MODULE__, [], name: via(ref))
+#     def append(ref, chunk), do: GenServer.cast(via(ref), {:append, chunk})
+#     def pop_chunk_with_overlap(ref, keep_ms \\ 1000), do: GenServer.call(via(ref), {:pop_chunk_with_overlap, keep_ms})
+#     def get_all(pid) do
+#         GenServer.call(pid, :get_all)
+#     end
+
+#     defp via(ref), do: {:via, Registry, {WhisperLive.AudioRegistry, ref}}
+
+#     @impl true
+#     def init(_), do: {:ok, %{buffer: <<>>, last_overlap: <<>>, rate: 48_000}}
+
+#     @impl true
+#     def handle_cast({:append, {_ts, chunk}}, state) do
+#         {:noreply, %{state | buffer: state.buffer <> chunk}}
+#     end
+
+#     @impl true
+#     def handle_call({:pop_chunk_with_overlap, keep_ms}, _from, state) do
+#         bytes_per_ms = div(state.rate * 2, 1000)
+#         total = state.last_overlap <> state.buffer
+
+#         total_size = byte_size(total)
+#         keep_bytes = keep_ms * bytes_per_ms
+#         overlap_bytes = min(keep_bytes, total_size)
+
+#         {to_send, to_keep} = split_bytes(total, overlap_bytes)
+
+#         new_state = %{state | buffer: <<>>, last_overlap: to_keep}
+#         {:reply, {to_send, state.rate}, new_state}
+#     end
+
+#     defp split_bytes(binary, keep_bytes) do
+#         total = byte_size(binary)
+#         send_bytes = max(total - keep_bytes, 0)
+#         <<send::binary-size(send_bytes), keep::binary>> = binary
+#         {send, keep}
+#     end
+
+
+#     def handle_call(:get_all, _from, state) do
+#         {:reply, Enum.reverse(state.buffer), state}
+#     end
+# end

whisper_live/lib/whisper_live/audio_full_buffer.ex

@@ -0,0 +1,17 @@
+defmodule WhisperLive.AudioFullBuffer do
+  use GenServer
+
+  def start_link(_), do: GenServer.start_link(__MODULE__, [], name: __MODULE__)
+  def append(chunk), do: GenServer.cast(__MODULE__, {:append, chunk})
+  def get_all(), do: GenServer.call(__MODULE__, :get_all)
+
+  def init(_), do: {:ok, %{chunks: [], rate: 48_000}}
+
+  def handle_cast({:append, {_rate, chunk}}, state) do
+    {:noreply, %{state | chunks: [chunk | state.chunks]}}
+  end
+
+  def handle_call(:get_all, _from, state) do
+    {:reply, {state.rate, Enum.reverse(state.chunks)}, state}
+  end
+end

whisper_live/lib/whisper_live/audio_stream_buffer.ex

@@ -0,0 +1,30 @@
+# mantiene un buffer de tiempo corto + overlap
+defmodule WhisperLive.AudioStreamBuffer do
+  use GenServer
+
+  def start_link(_), do: GenServer.start_link(__MODULE__, [], name: __MODULE__)
+  def append(chunk), do: GenServer.cast(__MODULE__, {:append, chunk})
+  def pop_chunk_with_overlap(n_seconds), do: GenServer.call(__MODULE__, {:pop_chunk, n_seconds})
+
+  def init(_), do: {:ok, %{buffer: <<>>, last_overlap: <<>>, rate: 48_000}}
+
+  def handle_cast({:append, {_rate, chunk}}, state) do
+    {:noreply, %{state | buffer: state.buffer <> chunk}}
+  end
+
+  def handle_call({:pop_chunk, n}, _from, state) do
+    rate = state.rate
+    bytes_per_second = rate * 2
+    bytes_to_take = n * bytes_per_second
+    <<chunk::binary-size(bytes_to_take), rest::binary>> = state.buffer
+
+    overlap_ms = 1000  # 1 segundo
+    overlap_bytes = div(overlap_ms * bytes_per_second, 1000)
+    new_overlap = binary_part(chunk, byte_size(chunk) - overlap_bytes, overlap_bytes)
+
+    reply = state.last_overlap <> chunk
+    new_buffer = new_overlap <> rest
+
+    {:reply, {rate, reply}, %{state | buffer: new_buffer, last_overlap: new_overlap}}
+  end
+end

whisper_live/lib/whisper_live/transcriber.ex

@@ -22,46 +22,71 @@ defmodule WhisperLive.Transcriber do
     end
 
     def handle_info(:transcribe, %{ref: ref} = state) do
-        case AudioBuffer.get_all(ref) do
-            [] ->
-            :noop
+        case AudioBuffer.get_tiny(ref) do
+            [] -> :noop
 
             [{rate, _} | _] = chunks ->
-            merged = chunks |> Enum.map(fn {_, bin} -> bin end) |> IO.iodata_to_binary()
-            tmpfile = tmp_path("realtime_#{ref}")
-            :ok = File.write!(tmpfile, encode_wav(merged, rate))
-
-            case send_to_whisper(tmpfile) do
+                merged = chunks |> Enum.map(fn {_, bin} -> bin end) |> IO.iodata_to_binary()
+                tmpfile = "tmp/rt_#{System.system_time(:millisecond)}.wav"
+                File.mkdir_p!("tmp")
+                File.write!(tmpfile, encode_wav(merged, rate))
+                case send_to_whisper(tmpfile) do
                 {:ok, response} ->
-                PubSub.broadcast(WhisperLive.PubSub, "transcription", {:transcription, response})
+                    PubSub.broadcast(WhisperLive.PubSub, "transcription", {:transcription, response})
 
                 {:error, reason} ->
-                Logger.warning("Realtime transcription error: #{inspect(reason)}")
-            end
+                    Logger.warning("Realtime transcription error: #{inspect(reason)}")
+                end
+
+                File.rm(tmpfile)
+
+                case GenServer.whereis({:via, Registry, {WhisperLive.AudioRegistry, ref}}) do
+                    pid when is_pid(pid) ->
+                        if Process.alive?(pid) do
+                            AudioBuffer.clear_tiny(ref)
+                        else
+                            Logger.debug("AudioBuffer #{inspect(ref)} no está vivo.")
+                        end
+
+                    _ ->
+                        Logger.debug("AudioBuffer #{inspect(ref)} no existe.")
+                end
 
-            File.rm(tmpfile)
         end
 
         schedule()
         {:noreply, state}
     end
 
+    # def handle_info(:transcribe, %{ref: ref} = state) do
+    #     case AudioBuffer.pop_chunk_with_overlap(ref, 1000) do
+    #         {"", _rate} ->
+    #         :noop
+
+    #         {audio, rate} ->
+    #         tmpfile = "tmp/rt_#{ref}_#{System.system_time(:millisecond)}.wav"
+    #         File.mkdir_p!("tmp")
+    #         File.write!(tmpfile, encode_wav(audio, rate))
+
+    #         case send_to_whisper(tmpfile) do
+    #             {:ok, response} ->
+    #             PubSub.broadcast(WhisperLive.PubSub, "transcription", {:transcription, response})
+
+    #             {:error, reason} ->
+    #             Logger.warning("Realtime transcription error: #{inspect(reason)}")
+    #         end
+    #     end
+
+    #     schedule()
+    #     {:noreply, state}
+    # end
+
     defp tmp_path(prefix) do
         unique = :erlang.unique_integer([:positive]) |> Integer.to_string()
         filename = prefix <> "_" <> unique <> ".wav"
         Path.join(System.tmp_dir!(), filename)
     end
 
-
-    # def handle_info({:transcription, raw_json}, socket) do
-    #     new_text =
-    #     raw_json
-    #     |> Jason.decode!()
-    #     |> get_in(["chunks", Access.at(0), "text"])
-
-    #     {:noreply, update(socket, :transcription, &(&1 <> " " <> new_text))}
-    # end
-
     defp schedule, do: Process.send_after(self(), :transcribe, @interval_ms)
 
     defp encode_wav(data, sample_rate) do

whisper_live/lib/whisper_live_web/channels/audio_channel.ex

@@ -7,10 +7,20 @@ defmodule WhisperLiveWeb.AudioChannel do
         ref = socket_id(socket)
         Logger.info("Cliente conectado al canal audio:lobby")
         {:ok, _} = AudioBuffer.start_link(ref)
+        # {:ok, _} = AudioFullBuffer.start_link(ref)
         {:ok, _} = WhisperLive.Transcriber.start_link(ref)
         {:ok, socket}
     end
 
+    # def handle_in("audio_chunk", %{"chunk" => chunk}, socket) do
+    #     decoded_chunk = Base.decode64!(chunk)
+
+    #     AudioStreamBuffer.append({48_000, decoded_chunk})
+    #     AudioFullBuffer.append({48_000, decoded_chunk})
+
+    #     {:noreply, socket}
+    # end
+
     def handle_in("audio_chunk", %{"data" => data, "sample_rate" => rate}, socket) do
         {:ok, binary} = Base.decode64(data)
         AudioBuffer.append(socket_id(socket), {rate, binary})
@@ -18,6 +28,29 @@ defmodule WhisperLiveWeb.AudioChannel do
         {:noreply, socket}
     end
 
+    # def handle_in("stop_audio", _payload, socket) do
+    #     Logger.info("🛑 Grabación detenida por cliente")
+
+    #     ref = socket_id(socket)
+
+    #     case AudioFullBuffer.get_all(ref) do
+    #     [{rate, _} | _] = chunks ->
+    #         merged = chunks |> Enum.map(fn {_, bin} -> bin end) |> IO.iodata_to_binary()
+    #         filename = "recordings/recording_#{System.system_time(:millisecond)}.wav"
+    #         File.mkdir_p!("recordings")
+    #         File.write!(filename, encode_wav(merged, rate))
+    #         whisper_large(filename)
+    #         File.rm!(filename)
+    #     _ ->
+    #         Logger.warning("No se recibieron chunks de audio")
+    #     end
+
+    #     AudioStreamBuffer.stop(ref)
+    #     AudioFullBuffer.stop(ref)
+    #     WhisperLive.Transcriber.stop(ref)
+    #     {:noreply, socket}
+    # end
+    
     def handle_in("stop_audio", _payload, socket) do
         Logger.info("🛑 Grabación detenida por cliente")
 
@@ -32,7 +65,7 @@ defmodule WhisperLiveWeb.AudioChannel do
             whisper_large(filename)
             File.rm!(filename)
         _ ->
-            Logger.warning("⚠️ No se recibieron chunks de audio")
+            Logger.warning("No se recibieron chunks de audio")
         end
 
         AudioBuffer.stop(ref)

whisper_live/lib/whisper_live_web/live/recorder.ex

@@ -65,135 +65,136 @@ defmodule WhisperLiveWeb.Live.Recorder do
                 </button>
             </div>
 
-            <div id="status" class="text-sm text-gray-600"></div>
+            <div id="status" class="hidden"></div>
 
-            <div id="transcriptionContainer" class="space-y-2">
-                <div class="p-2 bg-gray-100 rounded shadow">
-                <h2 class="text-sm font-semibold text-gray-700 mb-1">🟠 Transcripción en vivo</h2>
-                <p id="transcription" class="text-orange-600 whitespace-pre-wrap"><%= @transcription %></p>
-            </div>
-
-            <%= if @transcription_m != "" do %>
-                <div class="p-2 bg-gray-100 rounded shadow">
-                    <h2 class="text-sm font-semibold text-gray-700 mb-1">✅ Transcripción mejorada</h2>
-                    <p class="text-green-600 whitespace-pre-wrap"><%= @transcription_m %></p>
+            <div id="transcriptionContainer" class="w-full max-w-2xl space-y-4">
+                <div class="p-4 bg-gray-100 rounded shadow-md">
+                    <h2 class="text-sm font-semibold text-gray-700 mb-2">🟠 Transcripción en vivo</h2>
+                    <p id="transcription" class="text-orange-600 whitespace-pre-wrap break-words text-sm leading-relaxed"><%= @transcription %></p>
                 </div>
-            <% end %>
+
+                <%= if @transcription_m != "" do %>
+                    <div class="p-4 bg-gray-100 rounded shadow-md">
+                        <h2 class="text-sm font-semibold text-gray-700 mb-2">✅ Transcripción mejorada</h2>
+                        <p class="text-green-600 whitespace-pre-wrap break-words text-sm leading-relaxed"><%= @transcription_m %></p>
+                    </div>
+                <% end %>
             </div>
+
             <script type="module">
-            import { Socket } from "https://cdn.skypack.dev/phoenix"
+                import { Socket } from "https://cdn.skypack.dev/phoenix"
 
-            const startButton = document.getElementById("startButton")
-            const stopButton = document.getElementById("stopButton")
-            const statusDiv = document.getElementById("status")
+                const startButton = document.getElementById("startButton")
+                const stopButton = document.getElementById("stopButton")
+                const statusDiv = document.getElementById("status")
 
-            let socket = null
-            let channel = null
-            let audioContext = null
-            let processor = null
-            let mediaStream = null
-            let buffer = []
-            let sendInterval = null
+                let socket = null
+                let channel = null
+                let audioContext = null
+                let processor = null
+                let mediaStream = null
+                let buffer = []
+                let sendInterval = null
 
-            const sampleRate = 48000
+                const sampleRate = 48000
 
-            async function startRecording() {
-                startButton.disabled = true
-                stopButton.disabled = false
-                statusDiv.textContent = "🎙 Grabando..."
+                async function startRecording() {
+                    startButton.disabled = true
+                    stopButton.disabled = false
+                    statusDiv.textContent = "🎙 Grabando..."
 
-                socket = new Socket("ws://localhost:4004/socket")
-                socket.connect()
-                channel = socket.channel("audio:lobby")
+                    socket = new Socket("ws://localhost:4004/socket")
+                    socket.connect()
+                    channel = socket.channel("audio:lobby")
 
-                await channel.join()
-                .receive("ok", () => {
-                    console.log("✅ Canal conectado")
-                    statusDiv.textContent = "✅ Canal conectado"
-                })
-                .receive("error", () => {
-                    console.error("❌ Error al conectar canal")
-                    statusDiv.textContent = "❌ Error canal"
-                })
+                    await channel.join()
+                    .receive("ok", () => {
+                        console.log("✅ Canal conectado")
+                        statusDiv.textContent = "✅ Canal conectado"
+                    })
+                    .receive("error", () => {
+                        console.error("❌ Error al conectar canal")
+                        statusDiv.textContent = "❌ Error canal"
+                    })
 
-                try {
-                audioContext = new AudioContext({ sampleRate })
-                mediaStream = await navigator.mediaDevices.getUserMedia({ audio: true })
-                } catch (err) {
-                console.error("❌ Micrófono error:", err)
-                statusDiv.textContent = "❌ Error accediendo al micrófono"
-                return
-                }
-
-                const source = audioContext.createMediaStreamSource(mediaStream)
-                processor = audioContext.createScriptProcessor(4096, 1, 1)
-                source.connect(processor)
-                processor.connect(audioContext.destination)
-
-                buffer = []
-                processor.onaudioprocess = e => {
-                const input = e.inputBuffer.getChannelData(0)
-                const pcm = new Int16Array(input.length)
-                for (let i = 0; i < input.length; i++) {
-                    let s = Math.max(-1, Math.min(1, input[i]))
-                    pcm[i] = s < 0 ? s * 0x8000 : s * 0x7FFF
-                }
-                buffer.push(pcm)
-                }
-
-                sendInterval = setInterval(() => {
-                if (buffer.length === 0) return
-                const merged = flattenInt16(buffer)
-                buffer = []
-
-                function encodeBase64(uint8Array) {
-                    let binary = ''
-                    const len = uint8Array.byteLength
-                    for (let i = 0; i < len; i++) {
-                    binary += String.fromCharCode(uint8Array[i])
+                    try {
+                    audioContext = new AudioContext({ sampleRate })
+                    mediaStream = await navigator.mediaDevices.getUserMedia({ audio: true })
+                    } catch (err) {
+                    console.error("❌ Micrófono error:", err)
+                    statusDiv.textContent = "❌ Error accediendo al micrófono"
+                    return
                     }
-                    return btoa(binary)
+
+                    const source = audioContext.createMediaStreamSource(mediaStream)
+                    processor = audioContext.createScriptProcessor(4096, 1, 1)
+                    source.connect(processor)
+                    processor.connect(audioContext.destination)
+
+                    buffer = []
+                    processor.onaudioprocess = e => {
+                    const input = e.inputBuffer.getChannelData(0)
+                    const pcm = new Int16Array(input.length)
+                    for (let i = 0; i < input.length; i++) {
+                        let s = Math.max(-1, Math.min(1, input[i]))
+                        pcm[i] = s < 0 ? s * 0x8000 : s * 0x7FFF
+                    }
+                    buffer.push(pcm)
+                    }
+
+                    sendInterval = setInterval(() => {
+                        if (buffer.length === 0) return
+                        const merged = flattenInt16(buffer)
+                        buffer = []
+
+                        function encodeBase64(uint8Array) {
+                            let binary = ''
+                            const len = uint8Array.byteLength
+                            for (let i = 0; i < len; i++) {
+                            binary += String.fromCharCode(uint8Array[i])
+                            }
+                            return btoa(binary)
+                        }
+
+                        const base64 = encodeBase64(new Uint8Array(merged.buffer))
+                        channel.push("audio_chunk", { data: base64, sample_rate: sampleRate })
+                        console.log("Chunk enviado")
+                    }, 1000)
                 }
 
-                const base64 = encodeBase64(new Uint8Array(merged.buffer))
-                channel.push("audio_chunk", { data: base64, sample_rate: sampleRate })
-                console.log("📤 Enviado chunk")
-                }, 2000)
-            }
+                function stopRecording() {
+                    stopButton.disabled = true
+                    startButton.disabled = false
+                    statusDiv.textContent = "🛑 Grabación detenida."
 
-            function stopRecording() {
-                stopButton.disabled = true
-                startButton.disabled = false
-                statusDiv.textContent = "🛑 Grabación detenida."
+                    if (processor) processor.disconnect()
+                    if (audioContext) audioContext.close()
+                    if (mediaStream) mediaStream.getTracks().forEach(t => t.stop())
+                    if (sendInterval) clearInterval(sendInterval)
 
-                if (processor) processor.disconnect()
-                if (audioContext) audioContext.close()
-                if (mediaStream) mediaStream.getTracks().forEach(t => t.stop())
-                if (sendInterval) clearInterval(sendInterval)
-
-                if (channel) {
-                channel.push("stop_audio")
-                setTimeout(() => {
-                    channel.leave()
-                    socket.disconnect()
-                    console.log("🔌 Socket cerrado")
-                }, 500)
+                    if (channel) {
+                    channel.push("stop_audio")
+                    setTimeout(() => {
+                        channel.leave()
+                        socket.disconnect()
+                        console.log("🔌 Socket cerrado")
+                    }, 500)
+                    }
                 }
-            }
 
-            function flattenInt16(buffers) {
-                const length = buffers.reduce((acc, b) => acc + b.length, 0)
-                const out = new Int16Array(length)
-                let offset = 0
-                for (const b of buffers) {
-                out.set(b, offset)
-                offset += b.length
+                function flattenInt16(buffers) {
+                    const length = buffers.reduce((acc, b) => acc + b.length, 0)
+                    const out = new Int16Array(length)
+                    let offset = 0
+                    for (const b of buffers) {
+                    out.set(b, offset)
+                    offset += b.length
+                    }
+                    return out
                 }
-                return out
-            }
 
-            startButton.onclick = startRecording
-            stopButton.onclick = stopRecording
+                startButton.onclick = startRecording
+                stopButton.onclick = stopRecording
             </script>
         </div>
         """