# INSTRUMENT-AUSNAHME: Victor-Go — Soundscape: Titelmusik + Atmo + SFX für Holmes/Watson Hörspiel """ Baut watson_mathe_final.mp3: - Intro: Titelmusik 8s solo - Atmo: Kaminfeuer durchgehend bei -28dB - SFX: Papierrascheln, Uhrtickern, Federkiel — positions-genau eingebaut - Dialog: alle 53 Clips mit Pausen (PAUSE_AFTER) - Outro: Titelmusik zurück, 8s Ausklang Freesound API Key: fa1qGiJ4ugA1c8BRKogNKp4y5R1JM3wwNEqbzxDu """ import json, time, urllib.request, urllib.parse from pathlib import Path from pydub import AudioSegment, effects # ─── Pfade ─────────────────────────────────────────────────────────────────── CLIPS = Path("/Users/victorholland/Vibe Coding/dispatcher/cockpit/watson_demo_clips") SFX_DIR = Path("/Users/victorholland/Vibe Coding/dispatcher/cockpit/watson_sfx") TITLE_MUSIC = Path("/Users/victorholland/Desktop/Sherlock Holmes Titelmusik.mp3") OUT = CLIPS / "watson_mathe_final.mp3" LOG = Path("/tmp/watson_soundscape.log") SFX_DIR.mkdir(exist_ok=True) FREESOUND_KEY = "fa1qGiJ4ugA1c8BRKogNKp4y5R1JM3wwNEqbzxDu" DIALOG = [ ("d01","holmes"), ("d02","watson"), ("d03","holmes"), ("d04","watson"), ("d05","holmes"), ("d06","holmes"), ("d07","watson"), ("d08","holmes"), ("d09","watson"), ("d10","holmes"), ("d11","watson"), ("d12","holmes"), ("d13","watson"), ("d14","holmes"), ("d15","watson"), ("d16","holmes"), ("d17","holmes"), ("d18","watson"), ("d19","holmes"), ("d20","watson"), ("d21","holmes"), ("d22","watson"), ("d23","holmes"), ("d24","watson"), ("d25","holmes"), ("d26","watson"), ("d27","holmes"), ("d28","watson"), ("d29","holmes"), ("d30","watson"), ("d31","holmes"), ("d32","watson"), ("d33","holmes"), ("d34","watson"), ("d35","holmes"), ("d36","watson"), ("d37","holmes"), ("d38","watson"), ("d39","holmes"), ("d40","watson"), ("d41","holmes"), ("d42","watson"), ("d43","holmes"), ("d44","watson"), ("d45","holmes"), ("d46","watson"), ("d47","holmes"), ("d48","watson"), ("d49","holmes"), ("d50","watson"), ("d51","holmes"), ("d52","watson"), ("d53","holmes"), ] # Pausen nach bestimmten Clips (ms) PAUSE_AFTER = { "d01": 400, "d03": 300, "d05": 700, "d08": 600, "d10": 700, "d11": 200, "d13": 400, "d15": 200, "d16": 900, # Themenübergang Pythagoras "d17": 500, "d19": 800, "d21": 400, "d23": 600, "d25": 500, "d28": 900, # Themenübergang LGS "d29": 800, "d31": 400, "d33": 300, "d35": 700, # Themenübergang Textaufgaben "d37": 700, "d38": 300, "d39": 700, "d40": 900, "d41": 500, "d42": 700, "d43": 700, "d44": 300, "d45": 500, } # SFX-Positionen relativ zum Dialog-Start: {cue_label: (clip_id, offset_within_pause_ms, sfx_file)} # SFX werden in die jeweilige Pause NACH dem genannten Clip gelegt SFX_CUES = { "paper_at_start": ("d01", -800, "paper_rustle.mp3"), # 800ms vor d01-Ende = bei Papier "clock_thinking_1": ("d10", 150, "clock_ticking.mp3"), # kurz nach d10-Ende, Watson denkt "clock_thinking_2": ("d19", 200, "clock_ticking.mp3"), # Zirkus-Pause "clock_chapter_3": ("d28", 100, "clock_ticking.mp3"), # LGS-Übergang "pen_writing_1": ("d38", -600, "pen_writing.mp3"), # Watson rechnet in d38 "pen_writing_2": ("d40", -800, "pen_writing.mp3"), # Watson löst in d40 } LEAD_MS = 30 TAIL_MS = 200 def log(msg): print(msg, flush=True) with LOG.open("a") as f: f.write(msg + "\n") # ─── Freesound Download ─────────────────────────────────────────────────────── def freesound_search(query, dur_min=1, dur_max=60): url = (f"https://freesound.org/apiv2/search/text/" f"?query={urllib.parse.quote(query)}" f"&filter=duration:[{dur_min}%20TO%20{dur_max}]" f"&fields=id,name,duration,previews,avg_rating,num_downloads" f"&sort=downloads_desc" f"&page_size=5" f"&token={FREESOUND_KEY}") with urllib.request.urlopen(url, timeout=30) as r: data = json.loads(r.read()) return data.get("results", []) def download_sfx(query, out_name, dur_min=1, dur_max=60): out_path = SFX_DIR / out_name if out_path.exists(): log(f" ↩ {out_name} bereits vorhanden") return out_path results = freesound_search(query, dur_min, dur_max) if not results: log(f" ✗ Kein Ergebnis für: {query}") return None best = results[0] preview_url = best["previews"].get("preview-hq-mp3") or best["previews"].get("preview-lq-mp3") if not preview_url: log(f" ✗ Kein Preview für: {best['name']}") return None req = urllib.request.Request(preview_url, headers={"User-Agent": "WatsonHoerspiel/1.0"}) with urllib.request.urlopen(req, timeout=30) as r: out_path.write_bytes(r.read()) log(f" ✓ {out_name} [{best['name']}] {best['duration']:.1f}s ★{best.get('avg_rating',0):.1f}") return out_path # ─── SFX herunterladen ──────────────────────────────────────────────────────── log("=== Freesound SFX Download ===") sfx_queries = [ ("fireplace crackling indoor warm", "fireplace.mp3", 10, 120), ("paper rustle document single", "paper_rustle.mp3", 1, 6), ("clock ticking old antique", "clock_ticking.mp3", 3, 30), ("quill pen writing paper scratch", "pen_writing.mp3", 2, 10), ] for query, fname, dmin, dmax in sfx_queries: download_sfx(query, fname, dmin, dmax) time.sleep(0.3) # ─── SSML-Filter für Timestamps ────────────────────────────────────────────── def get_speech_bounds(ts_path, audio_len_ms): try: alignment = json.loads(ts_path.read_text()) chars = alignment.get("characters", []) starts = alignment.get("character_start_times_seconds", []) ends = alignment.get("character_end_times_seconds", []) if not chars: raise ValueError("leer") in_tag, real = False, [] for c, s, e in zip(chars, starts, ends): if c == '<': in_tag = True if not in_tag: real.append((c, s, e)) if c == '>': in_tag = False if not real: raise ValueError("keine Sprachzeichen") return int(real[0][1] * 1000), int(real[-1][2] * 1000) except Exception: return 30, audio_len_ms - 30 # ─── Voice-Track aufbauen + Timeline tracken ───────────────────────────────── log("\n=== Voice-Track ===") segments = [] # [(did, AudioSegment)] clip_start = {} # did → start_ms im fertigen Voice-Track pos = 0 for did, sp in DIALOG: mp3 = CLIPS / f"{did}_{sp}.mp3" ts = CLIPS / f"{did}_{sp}.timestamps.json" if not mp3.exists(): log(f" ✗ {did} fehlt") continue raw = AudioSegment.from_mp3(str(mp3)) norm = effects.normalize(raw, headroom=1.0) s_ms, e_ms = get_speech_bounds(ts, len(norm)) cut_start = max(0, s_ms - LEAD_MS) cut_end = min(len(norm), e_ms + TAIL_MS) if cut_end - cut_start < 200: cut_start, cut_end = 0, len(norm) clipped = norm[cut_start:cut_end] clip_start[did] = pos segments.append((did, clipped)) pos += len(clipped) pause_ms = PAUSE_AFTER.get(did, 0) if pause_ms > 0: silence = AudioSegment.silent(duration=pause_ms, frame_rate=44100) segments.append((f"{did}_pause", silence)) pos += pause_ms voice_dur = pos log(f"Voice-Track: {voice_dur/1000:.1f}s") # Voice-Track zusammensetzen voice_track = segments[0][1] for _, seg in segments[1:]: voice_track = voice_track.append(seg, crossfade=0) # ─── Titelmusik ────────────────────────────────────────────────────────────── log("\n=== Titelmusik ===") title = AudioSegment.from_mp3(str(TITLE_MUSIC)) title = title.set_frame_rate(44100).set_channels(2) INTRO_SOLO_MS = 8000 # Titelmusik solo INTRO_FADEOUT = 3000 # Fade-Out nach dem Solo OUTRO_FADEIN = 4000 # Fade-In am Ende OUTRO_DURATION = 9000 # Outro-Länge # Intro: erste 11s der Titelmusik (8s voll + 3s fade-out) intro = title[:INTRO_SOLO_MS + INTRO_FADEOUT].fade_out(INTRO_FADEOUT) # Outro: Titelmusik ab ~10s (um Intro-Wiederholung zu vermeiden), 9s fade-in + voll outro_src = title[8000:8000 + OUTRO_FADEIN + OUTRO_DURATION] if len(outro_src) < OUTRO_FADEIN + OUTRO_DURATION: outro_src = title[:OUTRO_FADEIN + OUTRO_DURATION] outro = outro_src.fade_in(OUTRO_FADEIN) INTRO_OFFSET = INTRO_SOLO_MS # Dialog startet hier # ─── Kaminfeuer-Atmo (looping) ─────────────────────────────────────────────── total_dur = INTRO_OFFSET + voice_dur + OUTRO_DURATION + 2000 fire_raw = SFX_DIR / "fireplace.mp3" if fire_raw.exists(): fire = AudioSegment.from_mp3(str(fire_raw)).set_frame_rate(44100).set_channels(2) while len(fire) < total_dur + 5000: fire = fire + fire fire = fire[:total_dur] - 28 # -28dB = dezentes Raumgeräusch log(f"Feuer: {len(fire)/1000:.1f}s @ -28dB") else: fire = AudioSegment.silent(duration=total_dur, frame_rate=44100) log(" ⚠ Kein Feuer-SFX") # ─── Gesamt-Canvas aufbauen ─────────────────────────────────────────────────── log("\n=== Mix ===") total = AudioSegment.silent(duration=total_dur, frame_rate=44100).set_channels(2) # Feuer-Bett total = total.overlay(fire, position=0) # Titelmusik-Intro title_mono = intro.set_channels(2) if intro.channels == 1 else intro total = total.overlay(title_mono, position=0) # Voice-Track voice_stereo = voice_track.set_channels(2) if voice_track.channels == 1 else voice_track total = total.overlay(voice_stereo, position=INTRO_OFFSET) # Titelmusik-Outro outro_stereo = outro.set_channels(2) if outro.channels == 1 else outro outro_pos = INTRO_OFFSET + voice_dur + 500 total = total.overlay(outro_stereo, position=outro_pos) # ─── SFX einbauen ──────────────────────────────────────────────────────────── def load_sfx(fname, gain_db=0): p = SFX_DIR / fname if not p.exists(): return None seg = AudioSegment.from_mp3(str(p)).set_frame_rate(44100).set_channels(2) return seg + gain_db sfx_paper = load_sfx("paper_rustle.mp3", -12) sfx_clock = load_sfx("clock_ticking.mp3", -18) sfx_pen = load_sfx("pen_writing.mp3", -14) def overlay_sfx(canvas, sfx, clip_id, offset_ms): if sfx is None or clip_id not in clip_start: return canvas pos = INTRO_OFFSET + clip_start[clip_id] + offset_ms pos = max(0, min(pos, len(canvas) - len(sfx))) return canvas.overlay(sfx, position=pos) if sfx_paper: # Papierrascheln kurz vor dem ersten Wort (Holmes hebt Brief auf) total = overlay_sfx(total, sfx_paper[:1200], "d01", -200) log(" SFX paper_rustle @ d01-200ms") if sfx_clock: short_tick = sfx_clock[:min(len(sfx_clock), 3500)] for cue_clip in ["d10", "d19", "d28"]: total = overlay_sfx(total, short_tick, cue_clip, 150) log(f" SFX clock @ {cue_clip}+150ms") if sfx_pen: short_pen = sfx_pen[:min(len(sfx_pen), 2500)] for cue_clip in ["d38", "d40"]: total = overlay_sfx(total, short_pen, cue_clip, 500) log(f" SFX pen @ {cue_clip}+500ms") # ─── Export ────────────────────────────────────────────────────────────────── log(f"\nGesamt: {len(total)/1000:.1f}s — Exportiere…") total.export(str(OUT), format="mp3", bitrate="320k", tags={"title":"Holmes erklärt Watson Mathematik — Final", "artist":"Watson Demo", "album":"Baker Street Hörspiel"}) log(f"Fertig: {OUT.stat().st_size//1024} KB") print(f"\n✓ {OUT.name} {OUT.stat().st_size//1024} KB {len(total)/1000:.1f}s")