Bugfix

Casted inf_steps to an integer, and changed the audio variable resulting from the pipe call into what AudioLDM writes in their code

app.py

@@ -18,15 +18,15 @@ def texttoaudio(prompt, neg_prompt, seed, inf_steps, guidance_scale):
     audio = pipe(
                  prompt, 
                  negative_prompt=neg_prompt,
-                 num_inference_steps=inf_steps,
+                 num_inference_steps=int(inf_steps),
                  guidance_scale=guidance_scale,
                  audio_length_in_s=5.0,
                  generator=generator.manual_seed(int(seed)),
-                ).audios[0]
+                )["audios"]
 
     # save the audio sample as a .wav file
     # scipy.io.wavfile.write("output.wav", rate=16000, data=audio)
-    return (16000, audio)
+    return (16000, audio[0])
 
 iface = gr.Interface(fn=texttoaudio, title="Prompt, Neg Prompt, Seed, Inf Steps, Guidance Scale", inputs=["text", "text", "number", "number", "number"], outputs="audio")