feat: Introduce voice model and TTS features, refactor agent claim fl…

package.json

@@ -14,6 +14,7 @@
     "@daily-co/daily-js": "^0.85.0",
     "@daily-co/daily-react": "^0.24.0",
     "@elevenlabs/react": "^0.11.0",
+    "@gradio/client": "^2.1.0",
     "@neynar/react": "^1.2.22",
     "@pigment-css/react": "^0.0.30",
     "@privy-io/react-auth": "^3.3.0",

src/app/api/tts/route.ts

@@ -0,0 +1,158 @@
+import { NextRequest, NextResponse } from "next/server";
+import { Client } from "@gradio/client";
+
+export const dynamic = "force-dynamic";
+
+const HF_BASE = "https://logeshmoltspaces-ultimate-rvc.hf.space";
+const FN_INDEX = 51; // /partial_34 – TTS pipeline
+
+/**
+ * Fire-and-forget queue job using raw SSE polling.
+ * Returns the audio URL from the first "process_completed" event.
+ */
+async function runGradioJob(
+  sessionHash: string,
+  fnIndex: number,
+  data: unknown[]
+): Promise<string | null> {
+  // 1 – Join the queue
+  const joinRes = await fetch(`${HF_BASE}/gradio_api/queue/join`, {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({
+      data,
+      event_data: null,
+      fn_index: fnIndex,
+      session_hash: sessionHash,
+      trigger_id: 381,
+    }),
+  });
+
+  if (!joinRes.ok) {
+    throw new Error(`queue/join failed ${joinRes.status}: ${await joinRes.text()}`);
+  }
+
+  // 2 – Stream the SSE response until "process_completed"
+  const streamRes = await fetch(
+    `${HF_BASE}/gradio_api/queue/data?session_hash=${sessionHash}`,
+    { headers: { Accept: "text/event-stream" } }
+  );
+
+  if (!streamRes.ok || !streamRes.body) {
+    throw new Error(`queue/data failed: ${streamRes.status}`);
+  }
+
+  const reader = streamRes.body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
+
+  while (true) {
+    const { value, done } = await reader.read();
+    if (done) break;
+    buffer += decoder.decode(value, { stream: true });
+    const lines = buffer.split("\n");
+    buffer = lines.pop() ?? "";
+
+    for (const line of lines) {
+      if (!line.startsWith("data:")) continue;
+      let msg: any;
+      try { msg = JSON.parse(line.slice(5).trim()); } catch { continue; }
+
+      if (msg.msg === "process_completed") {
+        if (msg.success === false) {
+          throw new Error(msg.output?.error ?? "Gradio processing failed");
+        }
+        const output = msg.output?.data?.[0];
+        let url: string | null =
+          typeof output === "string" ? output : (output?.url ?? null);
+        if (url && url.startsWith("/")) url = `${HF_BASE}${url}`;
+        return url;
+      }
+
+      if (msg.msg === "queue_full") {
+        throw new Error("Gradio queue is full – try again in a moment");
+      }
+    }
+  }
+
+  return null;
+}
+
+/**
+ * POST /api/tts
+ * Body: { text: string, voiceModel?: string, ttsVoice?: string }
+ * Returns: { status, audio_url }
+ */
+export async function POST(req: NextRequest) {
+  try {
+    const body = await req.json();
+    const {
+      text,
+      voiceModel = "mrkrabs",
+      ttsVoice = "en-US-ChristopherNeural",
+    } = body as { text: string; voiceModel?: string; ttsVoice?: string };
+
+    if (!text || text.trim().length === 0) {
+      return NextResponse.json(
+        { status: "error", message: "text is required" },
+        { status: 400 }
+      );
+    }
+
+    const sessionHash = Math.random().toString(36).slice(2, 13);
+
+    // ── Step 1: Wake the Space via @gradio/client (fn /partial_7)
+    // This populates the server-side rvc_model dropdown so fn_index 51
+    // can validate the model name. The client connects fresh and calls the
+    // no-arg /partial_7 endpoint which returns the current model lists.
+    try {
+      const client = await Client.connect("logeshmoltspaces/ultimate-rvc");
+      await client.predict("/partial_7", []);
+    } catch {
+      // Non-fatal — proceed; the Space may already be warm
+    }
+
+    // ── Step 2: Run the actual TTS + RVC pipeline (fn_index 51 = /partial_34)
+    const audioUrl = await runGradioJob(sessionHash, FN_INDEX, [
+      text,         // 0  tts_text
+      voiceModel,   // 1  rvc_model slug
+      ttsVoice,     // 2  edge-tts voice
+      0,            // 3  pitch
+      0,            // 4  filter_radius
+      0,            // 5  rms_mix_rate
+      0,            // 6  protect
+      0,            // 7  hop_length
+      "rmvpe",      // 8  f0_method
+      0.3,          // 9  crepe_hop_length
+      1,            // 10 f0_autotune
+      0.33,         // 11 f0_autotune_strength
+      false,        // 12 f0_vad
+      false,        // 13 split_audio
+      1,            // 14 batch_size
+      false,        // 15 clean_audio
+      155,          // 16 clean_strength
+      true,         // 17 export_format
+      0.7,          // 18 rms_mix_rate (secondary)
+      "contentvec", // 19 embedder_model
+      null,         // 20 embedder_model_custom
+      0,            // 21 sid
+      0,            // 22 batch_threshold
+      44100,        // 23 sample_rate
+      "mp3",        // 24 output_format
+      "",           // 25 extra
+    ]);
+
+    return NextResponse.json({
+      status: "done",
+      audio_url: audioUrl,
+      voice_model: voiceModel,
+      tts_voice: ttsVoice,
+    });
+  } catch (error: any) {
+    console.error("TTS route error:", error);
+    return NextResponse.json(
+      { status: "error", message: error.message ?? "Internal Server Error" },
+      { status: 500 }
+    );
+  }
+}